Datasets:

sswt

/

arxiv_sample_50K

like 0

SCAT Robust Selfsupervised Contrastive Learning via Adversarial Training for Text Classification ; Despite their promising performance across various natural language processing NLP tasks, current NLP systems are vulnerable to textual adversarial attacks. To defend against these attacks, most existing methods apply adversarial training by incorporating adversarial examples. However, these methods have to rely on groundtruth labels to generate adversarial examples, rendering it impractical for largescale model pretraining which is commonly used nowadays for NLP and many other tasks. In this paper, we propose a novel learning framework called SCAT Selfsupervised Contrastive Learning via Adversarial Training, which can learn robust representations without requiring labeled data. Specifically, SCAT modifies random augmentations of the data in a fully labelfree manner to generate adversarial examples. Adversarial training is achieved by minimizing the contrastive loss between the augmentations and their adversarial counterparts. We evaluate SCAT on two text classification datasets using two stateoftheart attack schemes proposed recently. Our results show that SCAT can not only train robust language models from scratch, but it can also significantly improve the robustness of existing pretrained language models. Moreover, to demonstrate its flexibility, we show that SCAT can also be combined with supervised adversarial training to further enhance model robustness.

Sequential Neural Barriers for Scalable Dynamic Obstacle Avoidance ; There are two major challenges for scaling up robot navigation around dynamic obstacles the complex interaction dynamics of the obstacles can be hard to model analytically, and the complexity of planning and control grows exponentially in the number of obstacles. Datadriven and learningbased methods are thus particularly valuable in this context. However, datadriven methods are sensitive to distribution drift, making it hard to train and generalize learned models across different obstacle densities. We propose a novel method for compositional learning of Sequential Neural Control Barrier models SNCBFs to achieve scalability. Our approach exploits an important observation the spatial interaction patterns of multiple dynamic obstacles can be decomposed and predicted through temporal sequences of states for each obstacle. Through decomposition, we can generalize control policies trained only with a small number of obstacles, to environments where the obstacle density can be 100x higher. We demonstrate the benefits of the proposed methods in improving dynamic collision avoidance in comparison with existing methods including potential fields, endtoend reinforcement learning, and modelpredictive control. We also perform hardware experiments and show the practical effectiveness of the approach in the supplementary video.

Engineering nonHermitian Second Order Topological Insulator in Quasicrystals ; NonHermitian topological phases have gained immense attention due to their potential to unlock novel features beyond Hermitian bounds. PTsymmetric Parity Timereversal symmetric nonHermitian models have been studied extensively over the past decade. In recent years, the topological properties of general nonHermitian models, regardless of the balance between gains and losses, have also attracted vast attention. Here we propose a nonHermitian secondorder topological SOT insulator that hosts gapless corner states on a twodimensional quasicrystalline lattice QL. We first construct a nonHermitian extension of the BernevigHughesZhang BHZ model on a QL generated by the AmmanBeenker AB tiling. This model has real spectra and supports helical edge states. Corner states emerge by adding a proper Wilson mass term that gaps out the edge states. We propose two variations of the mass term that result in fascinating characteristics. In the first variation, we obtain a purely real spectra for the secondorder topological phase. In the latter, we get a complex spectra with corner states localized at only two corners. Our findings pave a path to engineering exotic SOT phases where corner states can be localized at designated corners.

Discovering Hierarchical Achievements in Reinforcement Learning via Contrastive Learning ; Discovering achievements with a hierarchical structure on procedurally generated environments poses a significant challenge. This requires agents to possess a broad range of abilities, including generalization and longterm reasoning. Many prior methods are built upon modelbased or hierarchical approaches, with the belief that an explicit module for longterm planning would be beneficial for learning hierarchical achievements. However, these methods require an excessive amount of environment interactions or large model sizes, limiting their practicality. In this work, we identify that proximal policy optimization PPO, a simple and versatile modelfree algorithm, outperforms the prior methods with recent implementation practices. Moreover, we find that the PPO agent can predict the next achievement to be unlocked to some extent, though with low confidence. Based on this observation, we propose a novel contrastive learning method, called achievement distillation, that strengthens the agent's capability to predict the next achievement. Our method exhibits a strong capacity for discovering hierarchical achievements and shows stateoftheart performance on the challenging Crafter environment using fewer model parameters in a sampleefficient regime.

Blocks2World Controlling Realistic Scenes with Editable Primitives ; We present Blocks2World, a novel method for 3D scene rendering and editing that leverages a twostep process convex decomposition of images and conditioned synthesis. Our technique begins by extracting 3D parallelepipeds from various objects in a given scene using convex decomposition, thus obtaining a primitive representation of the scene. These primitives are then utilized to generate paired data through simple raytraced depth maps. The next stage involves training a conditioned model that learns to generate images from the 2Drendered convex primitives. This step establishes a direct mapping between the 3D model and its 2D representation, effectively learning the transition from a 3D model to an image. Once the model is fully trained, it offers remarkable control over the synthesis of novel and edited scenes. This is achieved by manipulating the primitives at test time, including translating or adding them, thereby enabling a highly customizable scene rendering process. Our method provides a fresh perspective on 3D scene rendering and editing, offering control and flexibility. It opens up new avenues for research and applications in the field, including authoring and data augmentation.

Robust Universal Inference ; In statistical inference, it is rarely realistic that the hypothesized statistical model is wellspecified, and consequently it is important to understand the effects of misspecification on inferential procedures. When the hypothesized statistical model is misspecified, the natural target of inference is a projection of the data generating distribution onto the model. We present a general method for constructing valid confidence sets for such projections, under weak regularity conditions, despite possible model misspecification. Our method builds upon the universal inference method of Wasserman et al. 2020 and is based on inverting a family of splitsample tests of relative fit. We study settings in which our methods yield either exact or approximate, finitesample valid confidence sets for various projection distributions. We study rates at which the resulting confidence sets shrink around the target of inference and complement these results with a simulation study.

Counterfactual Explanation for Fairness in Recommendation ; Fairnessaware recommendation eliminates discrimination issues to build trustworthy recommendation systems.Explaining the causes of unfair recommendations is critical, as it promotes fairness diagnostics, and thus secures users' trust in recommendation models. Existing fairness explanation methods suffer high computation burdens due to the largescale search space and the greedy nature of the explanation search process. Besides, they perform scorebased optimizations with continuous values, which are not applicable to discrete attributes such as gender and race. In this work, we adopt the novel paradigm of counterfactual explanation from causal inference to explore how minimal alterations in explanations change model fairness, to abandon the greedy search for explanations. We use realworld attributes from Heterogeneous Information Networks HINs to empower counterfactual reasoning on discrete attributes. We propose a novel Counterfactual Explanation for Fairness CFairER that generates attributelevel counterfactual explanations from HINs for recommendation fairness. Our CFairER conducts offpolicy reinforcement learning to seek highquality counterfactual explanations, with an attentive action pruning reducing the search space of candidate counterfactuals. The counterfactual explanations help to provide rational and proximate explanations for model fairness, while the attentive action pruning narrows the search space of attributes. Extensive experiments demonstrate our proposed model can generate faithful explanations while maintaining favorable recommendation performance.

Geometric Constraints in Probabilistic Manifolds A Bridge from Molecular Dynamics to Structured Diffusion Processes ; Understanding the macroscopic characteristics of biological complexes demands precision and specificity in statistical ensemble modeling. One of the primary challenges in this domain lies in sampling from particular subsets of the statespace, driven either by existing structural knowledge or specific areas of interest within the statespace. We propose a method that enables sampling from distributions that rigorously adhere to arbitrary sets of geometric constraints in Euclidean spaces. This is achieved by integrating a constraint projection operator within the wellregarded architecture of Denoising Diffusion Probabilistic Models, a framework founded in generative modeling and probabilistic inference. The significance of this work becomes apparent, for instance, in the context of deep learningbased drug design, where it is imperative to maintain specific molecular profile interactions to realize the desired therapeutic outcomes and guarantee safety.

Conformalization of Sparse Generalized Linear Models ; Given a sequence of observable variables x1, y1, ldots, xn, yn, the conformal prediction method estimates a confidence set for yn1 given xn1 that is valid for any finite sample size by merely assuming that the joint distribution of the data is permutation invariant. Although attractive, computing such a set is computationally infeasible in most regression problems. Indeed, in these cases, the unknown variable yn1 can take an infinite number of possible candidate values, and generating conformal sets requires retraining a predictive model for each candidate. In this paper, we focus on a sparse linear model with only a subset of variables for prediction and use numerical continuation techniques to approximate the solution path efficiently. The critical property we exploit is that the set of selected variables is invariant under a small perturbation of the input data. Therefore, it is sufficient to enumerate and refit the model only at the change points of the set of active features and smoothly interpolate the rest of the solution via a PredictorCorrector mechanism. We show how our pathfollowing algorithm accurately approximates conformal prediction sets and illustrate its performance using synthetic and real data examples.

Deep Probabilistic Movement Primitives with a Bayesian Aggregator ; Movement primitives are trainable parametric models that reproduce robotic movements starting from a limited set of demonstrations. Previous works proposed simple linear models that exhibited high sample efficiency and generalization power by allowing temporal modulation of movements reproducing movements faster or slower, blending merging two movements into one, viapoint conditioning constraining a movement to meet some particular viapoints and context conditioning generation of movements based on an observed variable, e.g., position of an object. Previous works have proposed neural networkbased motor primitive models, having demonstrated their capacity to perform tasks with some forms of input conditioning or timemodulation representations. However, there has not been a single unified deep motor primitive's model proposed that is capable of all previous operations, limiting neural motor primitive's potential applications. This paper proposes a deep movement primitive architecture that encodes all the operations above and uses a Bayesian context aggregator that allows a more sound context conditioning and blending. Our results demonstrate our approach can scale to reproduce complex motions on a larger variety of input choices compared to baselines while maintaining operations of linear movement primitives provide.

SleepEGAN A GANenhanced Ensemble Deep Learning Model for Imbalanced Classification of Sleep Stages ; Deep neural networks have played an important role in automatic sleep stage classification because of their strong representation and inmodel feature transformation abilities. However, class imbalance and individual heterogeneity which typically exist in raw EEG signals of sleep data can significantly affect the classification performance of any machine learning algorithms. To solve these two problems, this paper develops a generative adversarial network GANpowered ensemble deep learning model, named SleepEGAN, for the imbalanced classification of sleep stages. To alleviate class imbalance, we propose a new GAN called EGAN architecture adapted to the features of EEG signals for data augmentation. The generated samples for the minority classes are used in the training process. In addition, we design a costfree ensemble learning strategy to reduce the model estimation variance caused by the heterogeneity between the validation and test sets, so as to enhance the accuracy and robustness of prediction performance. We show that the proposed method can improve classification accuracy compared to several existing stateoftheart methods using three public sleep datasets.

Opening up ChatGPT Tracking openness, transparency, and accountability in instructiontuned text generators ; Large language models that exhibit instructionfollowing behaviour represent one of the biggest recent upheavals in conversational interfaces, a trend in large part fuelled by the release of OpenAI's ChatGPT, a proprietary large language model for text generation finetuned through reinforcement learning from human feedback LLMRLHF. We review the risks of relying on proprietary software and survey the first crop of opensource projects of comparable architecture and functionality. The main contribution of this paper is to show that openness is differentiated, and to offer scientific documentation of degrees of openness in this fastmoving field. We evaluate projects in terms of openness of code, training data, model weights, RLHF data, licensing, scientific documentation, and access methods. We find that while there is a fastgrowing list of projects billing themselves as 'open source', many inherit undocumented data of dubious legality, few share the allimportant instructiontuning a key site where human annotation labour is involved, and careful scientific documentation is exceedingly rare. Degrees of openness are relevant to fairness and accountability at all points, from data collection and curation to model architecture, and from training and finetuning to release and deployment.

Navigating the Complexity of Generative AI Adoption in Software Engineering ; In this paper, the adoption patterns of Generative Artificial Intelligence AI tools within software engineering are investigated. Influencing factors at the individual, technological, and societal levels are analyzed using a mixedmethods approach for an extensive comprehension of AI adoption. An initial structured interview was conducted with 100 software engineers, employing the Technology Acceptance Model TAM, the Diffusion of Innovations theory DOI, and the Social Cognitive Theory SCT as guiding theories. A theoretical model named the HumanAI Collaboration and Adaptation Framework HACAF was deduced using the Gioia Methodology, characterizing AI adoption in software engineering. This model's validity was subsequently tested through Partial Least Squares Structural Equation Modeling PLSSEM, using data collected from 183 software professionals. The results indicate that the adoption of AI tools in these early integration stages is primarily driven by their compatibility with existing development workflows. This finding counters the traditional theories of technology acceptance. Contrary to expectations, the influence of perceived usefulness, social aspects, and personal innovativeness on adoption appeared to be less significant. This paper yields significant insights for the design of future AI tools and supplies a structure for devising effective strategies for organizational implementation.

Tensor Decompositions Meet Control Theory Learning General Mixtures of Linear Dynamical Systems ; Recently Chen and Poor initiated the study of learning mixtures of linear dynamical systems. While linear dynamical systems already have wideranging applications in modeling timeseries data, using mixture models can lead to a better fit or even a richer understanding of underlying subpopulations represented in the data. In this work we give a new approach to learning mixtures of linear dynamical systems that is based on tensor decompositions. As a result, our algorithm succeeds without strong separation conditions on the components, and can be used to compete with the Bayes optimal clustering of the trajectories. Moreover our algorithm works in the challenging partiallyobserved setting. Our starting point is the simple but powerful observation that the classic HoKalman algorithm is a close relative of modern tensor decomposition methods for learning latent variable models. This gives us a playbook for how to extend it to work with more complicated generative models.

Explaining and visualizing blackbox models through counterfactual paths ; Explainable AI XAI is an increasingly important area of machine learning research, which aims to make blackbox models transparent and interpretable. In this paper, we propose a novel approach to XAI that uses the socalled counterfactual paths generated by conditional permutations of features. The algorithm measures feature importance by identifying sequential permutations of features that most influence changes in model predictions. It is particularly suitable for generating explanations based on counterfactual paths in knowledge graphs incorporating domain knowledge. Counterfactual paths introduce an additional graph dimension to current XAI methods in both explaining and visualizing blackbox models. Experiments with synthetic and medical data demonstrate the practical applicability of our approach.

Decoupled Quark Stars Relativistic Models in the Regime of Selfinteracting BransDicke Gravity ; In this work, we generate two static anisotropic solutions for a sphere containing quark matter in the framework of selfinteracting BransDicke theory. For this purpose, we add an anisotropic source in the seed distribution and decouple the field equations through deformation in the radial metric function. As a result of this transformation, the field equations are disintegrated into two systems which separately include the effects of isotropic and anisotropic sources. The system related to the additional source is solved via the MIT bag model equation of state. We consider Tolman V spacetime and Karmarkar condition to formulate two solutions for the isotropic sector which are extended to the anisotropic domain via decoupling technique. The junction conditions at the boundary determine the unknown parameters in terms of mass and radius of the spherical object. We investigate the viability and stability of the constructed strange star models in the presence of massive scalar field corresponding to three strange star candidates Her X1, PSR J16142230 and 4U 160852. It is concluded that the anisotropic models are wellbehaved as they fulfill the necessary requirements for lower as well as higher values of the decoupling parameter.

SLMGAN Exploiting Speech Language Model Representations for Unsupervised ZeroShot Voice Conversion in GANs ; In recent years, largescale pretrained speech language models SLMs have demonstrated remarkable advancements in various generative speech modeling applications, such as texttospeech synthesis, voice conversion, and speech enhancement. These applications typically involve mapping text or speech inputs to pretrained SLM representations, from which target speech is decoded. This paper introduces a new approach, SLMGAN, to leverage SLM representations for discriminative tasks within the generative adversarial network GAN framework, specifically for voice conversion. Building upon StarGANv2VC, we add our novel SLMbased WavLM discriminators on top of the melbased discriminators along with our newly designed SLM feature matching loss function, resulting in an unsupervised zeroshot voice conversion system that does not require text labels during training. Subjective evaluation results show that SLMGAN outperforms existing stateoftheart zeroshot voice conversion models in terms of naturalness and achieves comparable similarity, highlighting the potential of SLMbased discriminators for related applications.

Exact holeinduced SUN flavorsinglets in certain Uinfty SUN Hubbard models ; We prove that the motion of a single hole induces SUN flavorsinglets in the Uinfty SUN Fermi Hubbard model on a Husimilike tree graph. The result is also generalized to certain tJ models with antiferromagnetic interactions and singlet hopping terms typically neglected in the literature. This is an SUN generalization of the counterNagaoka theorem introduced in Phys. Rev. B 107, L140401 2023. Our results suggest the existence of antiferromagnetic or resonatingvalencebond RVBlike polarons in the tJ models on a more realistic nonbipartite lattice. Such antiferromagneticRVB polarons may be relevant for a novel strongcoupling mechanism of superconductivity or other exotic fractionalized phases of matter.

On the Origin of LLMs An Evolutionary Tree and Graph for 15,821 Large Language Models ; Since late 2022, Large Language Models LLMs have become very prominent with LLMs like ChatGPT and Bard receiving millions of users. Hundreds of new LLMs are announced each week, many of which are deposited to Hugging Face, a repository of machine learning models and datasets. To date, nearly 16,000 Text Generation models have been uploaded to the site. Given the huge influx of LLMs, it is of interest to know which LLM backbones, settings, training methods, and families are popular or trending. However, there is no comprehensive index of LLMs available. We take advantage of the relatively systematic nomenclature of Hugging Face LLMs to perform hierarchical clustering and identify communities amongst LLMs using ngrams and term frequencyinverse document frequency. Our methods successfully identify families of LLMs and accurately cluster LLMs into meaningful subgroups. We present a public web application to navigate and explore Constellation, our atlas of 15,821 LLMs. Constellation rapidly generates a variety of visualizations, namely dendrograms, graphs, word clouds, and scatter plots. Constellation is available at the following link httpsconstellation.sites.stanford.edu.

Large Language Models can accomplish Business Process Management Tasks ; Business Process Management BPM aims to improve organizational activities and their outcomes by managing the underlying processes. To achieve this, it is often necessary to consider information from various sources, including unstructured textual documents. Therefore, researchers have developed several BPMspecific solutions that extract information from textual documents using Natural Language Processing techniques. These solutions are specific to their respective tasks and cannot accomplish multiple processrelated problems as a generalpurpose instrument. However, in light of the recent emergence of Large Language Models LLMs with remarkable reasoning capabilities, such a generalpurpose instrument with multiple applications now appears attainable. In this paper, we illustrate how LLMs can accomplish textrelated BPM tasks by applying a specific LLM to three exemplary tasks mining imperative process models from textual descriptions, mining declarative process models from textual descriptions, and assessing the suitability of process tasks from textual descriptions for robotic process automation. We show that, without extensive configuration or prompt engineering, LLMs perform comparably to or better than existing solutions and discuss implications for future BPM research as well as practical usage.

Leveraging Plasmonic Hot Electrons to Quench Defect Emission in Metal Semiconductor Nanostructured Hybrids Experiment and Modeling ; Modeling lightmatter interaction in hybrid plasmonic materials is vital to their widening relevance from optoelectronics to photocatalysis. Here, we explore photoluminescence from ZnO nanorods ZNR embedded with gold nanoparticles Au NPs. A progressive increase in Au NP concentration introduces significant structural disorder and defects in the ZNRs, which paradoxically quenches defect related visible photoluminescence PL while intensifying the near band edge NBE emission. Under UV excitation, the simulated semiclassical model realizes PL from ZnO with subband gap defect states, eliciting visible emissions that are absorbed by Au NPs to generate a nonequilibrium hot carrier distribution. The photostimulated hot carriers, transferred to ZnO, substantially modify its steadystate luminescence, reducing NBE emission lifetime and altering the abundance of ionized defect states, finally reducing visible emission. The simulations show that the change in the interfacial band bending at the AuZnO interface under optical illumination facilitates charge transfer between the components. This work provides a general foundation to observe and model the hot carrier dynamics in hybrid plasmonic systems.

Discovering Active Subspaces for HighDimensional Computer Models ; Dimension reduction techniques have long been an important topic in statistics, and active subspaces AS have received much attention this past decade in the computer experiments literature. The most common approach towards estimating the AS is to use Monte Carlo with numerical gradient evaluation. While sensible in some settings, this approach has obvious drawbacks. Recent research has demonstrated that active subspace calculations can be obtained in closed form, conditional on a Gaussian process GP surrogate, which can be limiting in highdimensional settings for computational reasons. In this paper, we produce the relevant calculations for a more general case when the model of interest is a linear combination of tensor products. These general equations can be applied to the GP, recovering previous results as a special case, or applied to the models constructed by other regression techniques including multivariate adaptive regression splines MARS. Using a MARS surrogate has many advantages including improved scaling, better estimation of active subspaces in high dimensions and the ability to handle a large number of prior distributions in closed form. In one realworld example, we obtain the active subspace of a radiationtransport code with 240 inputs and 9,372 model runs in under half an hour.

Joint onesided synthetic unpaired image translation and segmentation for colorectal cancer prevention ; Deep learning has shown excellent performance in analysing medical images. However, datasets are difficult to obtain due privacy issues, standardization problems, and lack of annotations. We address these problems by producing realistic synthetic images using a combination of 3D technologies and generative adversarial networks. We propose CUTseg, a joint training where a segmentation model and a generative model are jointly trained to produce realistic images while learning to segment polyps. We take advantage of recent onesided translation models because they use significantly less memory, allowing us to add a segmentation model in the training loop. CUTseg performs better, is computationally less expensive, and requires less real images than other memoryintensive image translation approaches that require two stage training. Promising results are achieved on five real polyp segmentation datasets using only one real image and zero real annotations. As a part of this study we release SynthColon, an entirely synthetic dataset that includes 20000 realistic colon images and additional details about depth and 3D geometry httpsenric1994.github.iosynthcolon

Towards NonParametric Models for Confidence Aware Image Prediction from Low Data using Gaussian Processes ; The ability to envision future states is crucial to informed decision making while interacting with dynamic environments. With cameras providing a prevalent and information rich sensing modality, the problem of predicting future states from image sequences has garnered a lot of attention. Current state of the art methods typically train large parametric models for their predictions. Though often able to predict with accuracy, these models rely on the availability of large training datasets to converge to useful solutions. In this paper we focus on the problem of predicting future images of an image sequence from very little training data. To approach this problem, we use nonparametric models to take a probabilistic approach to image prediction. We generate probability distributions over sequentially predicted images and propagate uncertainty through time to generate a confidence metric for our predictions. Gaussian Processes are used for their data efficiency and ability to readily incorporate new training data online. We showcase our method by successfully predicting future frames of a smooth fluid simulation environment.

Unlocking ensemble ecosystem modelling for large and complex networks ; The potential effects of conservation actions on threatened species can be predicted using ensemble ecosystem models. These model ensembles commonly assume stable coexistence of species in the absence of available data. However, existing ensemblegeneration methods become computationally inefficient as the size of the ecosystem network increases, preventing larger networks from being studied. We present a novel sequential Monte Carlo sampling approach for ensemble generation that is orders of magnitude faster than existing approaches. We demonstrate that the methods produce equivalent parameter inferences, model predictions, and tightly constrained parameter combinations using a novel sensitivity analysis method. For one case study, we demonstrate a speedup from 108 days to 6 hours, while maintaining equivalent ensembles. Now, for the first time, larger and more realistic networks can be practically simulated.

Why Is Prompt Tuning for VisionLanguage Models Robust to Noisy Labels ; Visionlanguage models such as CLIP learn a generic textimage embedding from largescale training data. A visionlanguage model can be adapted to a new classification task through fewshot prompt tuning. We find that such a prompt tuning process is highly robust to label noises. This intrigues us to study the key reasons contributing to the robustness of the prompt tuning paradigm. We conducted extensive experiments to explore this property and find the key factors are 1 the fixed classname tokens provide a strong regularization to the optimization of the model, reducing gradients induced by the noisy samples; 2 the powerful pretrained imagetext embedding that is learned from diverse and generic web data provides strong prior knowledge for image classification. Further, we demonstrate that noisy zeroshot predictions from CLIP can be used to tune its own prompt, significantly enhancing prediction accuracy in the unsupervised setting. The code is available at httpsgithub.comCEWuPTNL.

Modelling functionalized drug release for a spherical capsule ; Advances in material design has led to the rapid development of novel materials with increasing complexity and functions in bioengineering. In particular, functionally graded materials FGMs offer important advantages in various fields of application. In this work, we consider a heterogeneous reactiondiffusion model for an FGM spherical drug releasing system that generalizes the multilayer configuration to arbitrary spatiallyvariable coefficients. Our model proposes a possible form for the drug diffusivity and reaction rate functions exhibiting fixed average material properties and a drug release profile that can be continuously varied between the limiting cases of a homogeneous system constant coefficients and twolayer system stepwise coefficients. A hybrid analyticalnumerical solution is then used to solve the model, which provides closedform expressions for the drug concentration and drug release profiles in terms of generalized Fourier series. The resulting concentration and mass profiles show how the release rate can be controlled and continuously varied between a fast homogeneous and slow twolayer release.

Fashion Matrix Editing Photos by Just Talking ; The utilization of Large Language Models LLMs for the construction of AI systems has garnered significant attention across diverse fields. The extension of LLMs to the domain of fashion holds substantial commercial potential but also inherent challenges due to the intricate semantic interactions in fashionrelated generation. To address this issue, we developed a hierarchical AI system called Fashion Matrix dedicated to editing photos by just talking. This system facilitates diverse promptdriven tasks, encompassing garment or accessory replacement, recoloring, addition, and removal. Specifically, Fashion Matrix employs LLM as its foundational support and engages in iterative interactions with users. It employs a range of Semantic Segmentation Models e.g., GroundedSAM, MattingAnything, etc. to delineate the specific editing masks based on user instructions. Subsequently, Visual Foundation Models e.g., Stable Diffusion, ControlNet, etc. are leveraged to generate edited images from text prompts and masks, thereby facilitating the automation of fashion editing processes. Experiments demonstrate the outstanding ability of Fashion Matrix to explores the collaborative potential of functionally diverse pretrained models in the domain of fashion editing.

Compact Stars with Dark Energy in General Relativity and Modified Gravity ; We investigate realistic models of compact objects, focusing on neutron and strange stars, composed by dense matter and dark energy in the form of a simple fluid or scalar field interacting with matter. For the dark energy component, we use equations of state compatible with cosmological observations. This requirement strongly constrains possible deviations from the simple LambdaCold DarkMatter model with EoS pdrhod at least for small densities of the dark component. But we can propose that the density of dark energy interacting with matter can reach large values in relativistic stars and affects the star parameters such as the mass and radius. Simple models of dark energy are considered. Then we investigated possible effects from modified gravity choosing to study the R2 model combined with dark energy. Finally, the case of dark energy as scalar field nonminimally interacting with gravity is considered.

Macroscopic limit of a FokkerPlanck model of swarming rigid bodies ; We consider selfpropelled rigidbodies interacting through local bodyattitude alignment modelled by stochastic differential equations. We derive a hydrodynamic model of this system at large spatiotemporal scales and particle numbers in any dimension n geq 3. This goal was already achieved in dimension n3, or in any dimension n geq 3 for a different system involving jump processes. However, the present work corresponds to huge conceptual and technical gaps compared with earlier ones. The key difficulty is to determine an auxiliary but essential object, the generalized collision invariant. We achieve this aim by using the geometrical structure of the rotation group, namely, its maximal torus, Cartan subalgebra and Weyl group as well as other concepts of representation theory and Weyl's integration formula. The resulting hydrodynamic model appears as a hyperbolic system whose coefficients depend on the generalized collision invariant.

SAFE SaliencyAware Counterfactual Explanations for DNNbased Automated Driving Systems ; A CF explainer identifies the minimum modifications in the input that would alter the model's output to its complement. In other words, a CF explainer computes the minimum modifications required to cross the model's decision boundary. Current deep generative CF models often work with userselected features rather than focusing on the discriminative features of the blackbox model. Consequently, such CF examples may not necessarily lie near the decision boundary, thereby contradicting the definition of CFs. To address this issue, we propose in this paper a novel approach that leverages saliency maps to generate more informative CF explanations. Source codes are available at httpsgithub.comAmirSamadiSaliencyAwareCF.

LPMusicCaps LLMBased Pseudo Music Captioning ; Automatic music captioning, which generates natural language descriptions for given music tracks, holds significant potential for enhancing the understanding and organization of large volumes of musical data. Despite its importance, researchers face challenges due to the costly and timeconsuming collection process of existing musiclanguage datasets, which are limited in size. To address this data scarcity issue, we propose the use of large language models LLMs to artificially generate the description sentences from largescale tag datasets. This results in approximately 2.2M captions paired with 0.5M audio clips. We term it Large Language Model based Pseudo music caption dataset, shortly, LPMusicCaps. We conduct a systemic evaluation of the largescale music captioning dataset with various quantitative evaluation metrics used in the field of natural language processing as well as human evaluation. In addition, we trained a transformerbased music captioning model with the dataset and evaluated it under zeroshot and transferlearning settings. The results demonstrate that our proposed approach outperforms the supervised baseline model.

CDARL Contrastive diffusion adversarial representation learning for labelfree blood vessel segmentation ; Blood vessel segmentation in medical imaging is one of the essential steps for vascular disease diagnosis and interventional planning in a broad spectrum of clinical scenarios in imagebased medicine and interventional medicine. Unfortunately, manual annotation of the vessel masks is challenging and resourceintensive due to subtle branches and complex structures. To overcome this issue, this paper presents a selfsupervised vessel segmentation method, dubbed the contrastive diffusion adversarial representation learning CDARL model. Our model is composed of a diffusion module and a generation module that learns the distribution of multidomain blood vessel data by generating synthetic vessel images from diffusion latent. Moreover, we employ contrastive learning through a maskbased contrastive loss so that the model can learn more realistic vessel representations. To validate the efficacy, CDARL is trained using various vessel datasets, including coronary angiograms, abdominal digital subtraction angiograms, and retinal imaging. Experimental results confirm that our model achieves performance improvement over baseline methods with noise robustness, suggesting the effectiveness of CDARL for vessel segmentation.

Advancing Beyond Identification Multibit Watermark for Large Language Models ; We propose a method to tackle misuses of large language models beyond the identification of machinegenerated text. While existing methods focus on detection, some malicious misuses demand tracing the adversary user for counteracting them. To address this, we propose Multibit Watermark via Position Allocation, embedding traceable multibit information during language model generation. Leveraging the benefits of zerobit watermarking, our method enables robust extraction of the watermark without any model access, embedding and extraction of long messages geq 32bit without finetuning, and maintaining text quality, while allowing zerobit detection all at the same time. Moreover, our watermark is relatively robust under strong attacks like interleaving human texts and paraphrasing.

Graph Embedding Dynamic Featurebased Supervised Contrastive Learning of Transient Stability for Changing Power Grid Topologies ; Accurate online transient stability prediction is critical for ensuring power system stability when facing disturbances. While traditional transient stablity analysis replies on the time domain simulations can not be quickly adapted to the power grid toplogy change. In order to vectorize highdimensional power grid topological structure information into lowdimensional nodebased graph embedding streaming data, graph embedding dynamic feature GEDF has been proposed. The transient stability GEDFbased supervised contrastive learning GEDFSCL model uses supervised contrastive learning to predict transient stability with GEDFs, considering power grid topology information. To evaluate the performance of the proposed GEDFSCL model, power grids of varying topologies were generated based on the IEEE 39bus system model. Transient operational data was obtained by simulating N1 and Nbmm1 contingencies on these generated power system topologies. Test result demonstrated that the GEDFSCL model can achieve high accuracy in transient stability prediction and adapt well to changing power grid topologies.

Gravitationally induced matter creation and cosmological consequences ; In this work, a twofluid interacting model in a flat FLRW universe has been studied considering particle creation mechanism with a particular form of particle creation rate GammaGamma0 HfracGamma1H from different aspects. Statistical analysis with a combined data set of SNe Ia Supernovae Type Ia and Hubble data is performed to achieve the bestfit values of the model parameters, and the model is compatible with current observational data. We also perform a dynamical analysis of this model to get an overall qualitative description of the cosmological evolution by converting the governing equations into a system of ordinary differential equations considering a proper transformation of variables. We find some nonisolated sets of critical points, among which some usually are normally hyperbolic sets of points that describe the present acceleration of the Universe dominated by dark energy mimicking cosmological constant or phantom fluid. Scaling solutions are also obtained from this analysis, and they can alleviate the coincidence problem successfully. Finally, the thermodynamic analysis shows that the Generalized second law of thermodynamics is valid in an irreversible thermodynamic context.

Simulationbased inference using surjective sequential neural likelihood estimation ; We present Surjective Sequential Neural Likelihood SSNL estimation, a novel method for simulationbased inference in models where the evaluation of the likelihood function is not tractable and only a simulator that can generate synthetic data is available. SSNL fits a dimensionalityreducing surjective normalizing flow model and uses it as a surrogate likelihood function which allows for conventional Bayesian inference using either Markov chain Monte Carlo methods or variational inference. By embedding the data in a lowdimensional space, SSNL solves several issues previous likelihoodbased methods had when applied to highdimensional data sets that, for instance, contain noninformative data dimensions or lie along a lowerdimensional manifold. We evaluate SSNL on a wide variety of experiments and show that it generally outperforms contemporary methods used in simulationbased inference, for instance, on a challenging realworld example from astrophysics which models the magnetic field strength of the sun using a solar dynamo model.

Grounded Image Text Matching with Mismatched Relation Reasoning ; This paper introduces Grounded Image Text Matching with Mismatched Relation GITMMR, a novel visuallinguistic joint task that evaluates the relation understanding capabilities of transformerbased pretrained models. GITMMR requires a model to first determine if an expression describes an image, then localize referred objects or ground the mismatched parts of the text. We provide a benchmark for evaluating pretrained models on this task, with a focus on the challenging settings of limited data and outofdistribution sentence lengths. Our evaluation demonstrates that pretrained models lack data efficiency and length generalization ability. To address this, we propose the Relationsensitive Correspondence Reasoning Network RCRN, which incorporates relationaware reasoning via bidirectional message propagation guided by language structure. RCRN can be interpreted as a modular program and delivers strong performance in both length generalization and data efficiency.

Ultraviolet Running Constraints on Low Mass Dark Sectors ; We analyze the UV breakdown of SubGeV dark matter models that live in a new, dark U1 sector. Many of these models include a scalar field, which is either the dark matter itself or a dark Higgs field that generates mass terms for the dark matter particle via spontaneous symmetry breaking. A quartic self coupling of this scalar field is generically allowed, and we show that its running is largely governed by the strength of the U1 gauge field, alphaD. Furthermore, it consistently has a lower Landau pole than the gauge coupling. Link fields, which couple to both the dark sector and the Standard Model SM, connect these Landau poles to constraints on SM charged particles. Current LHC constraints on link fields are compatible with alphaD lesssim 0.5 1 for most of the mass range in most models, while smaller values, alphaD lesssim 0.15, are favored for Majorana DM.

DiffColor Toward High Fidelity TextGuided Image Colorization with Diffusion Models ; Recent datadriven image colorization methods have enabled automatic or referencebased colorization, while still suffering from unsatisfactory and inaccurate objectlevel color control. To address these issues, we propose a new method called DiffColor that leverages the power of pretrained diffusion models to recover vivid colors conditioned on a prompt text, without any additional inputs. DiffColor mainly contains two stages colorization with generative color prior and incontext controllable colorization. Specifically, we first finetune a pretrained texttoimage model to generate colorized images using a CLIPbased contrastive loss. Then we try to obtain an optimized text embedding aligning the colorized image and the text prompt, and a finetuned diffusion model enabling highquality image reconstruction. Our method can produce vivid and diverse colors with a few iterations, and keep the structure and background intact while having colors wellaligned with the target language guidance. Moreover, our method allows for incontext colorization, i.e., producing different colorization results by modifying prompt texts without any finetuning, and can achieve objectlevel controllable colorization results. Extensive experiments and user studies demonstrate that DiffColor outperforms previous works in terms of visual quality, color fidelity, and diversity of colorization options.

Optimizing Sorting of MicroSized BioCells in Symmetric Serpentine Microchannel using Machine Learning ; Efficient sorting of target cells is crucial for advancing cellular research in biology and medical diagnostics. Inertial microfluidics, an emerging technology, offers a promising approach for labelfree particle sorting with high throughput. This paper presents a comprehensive study employing numerical computational fluid dynamics CFD simulations to investigate particle migration and sorting within a symmetric serpentine microchannel. By adopting a Eulerian approach to solve fluid dynamics and a Lagrangian framework to track particles, the research explores the impact of flow Reynolds number and the number of loops in the serpentine channel on sorting efficiency. To generate a robust datadriven model, the authors performed CFD simulations for 200 combinations of randomly generated data points. The study leverages the collected data to develop a datacentric machine learning model capable of accurately predicting flow parameters for specific sorting efficiencies. Remarkably, the developed model achieved a 92 accuracy in predicting the Channel Reynolds Number during testing. However, it is worth noting that the model currently faces challenges in accurately predicting the required number of loops for efficient sorting.

Is GPT4 a reliable rater Evaluating Consistency in GPT4 Text Ratings ; This study investigates the consistency of feedback ratings generated by OpenAI's GPT4, a stateoftheart artificial intelligence language model, across multiple iterations, time spans and stylistic variations. The model rated responses to tasks within the Higher Education HE subject domain of macroeconomics in terms of their content and style. Statistical analysis was conducted in order to learn more about the interrater reliability, consistency of the ratings across iterations and the correlation between ratings in terms of content and style. The results revealed a high interrater reliability with ICC scores ranging between 0.94 and 0.99 for different timespans, suggesting that GPT4 is capable of generating consistent ratings across repetitions with a clear prompt. Style and content ratings show a high correlation of 0.87. When applying a nonadequate style the average content ratings remained constant, while style ratings decreased, which indicates that the large language model LLM effectively distinguishes between these two criteria during evaluation. The prompt used in this study is furthermore presented and explained. Further research is necessary to assess the robustness and reliability of AI models in various use cases.

PreTrained Large Language Models for Industrial Control ; For industrial control, developing highperformance controllers with few samples and low technical debt is appealing. Foundation models, possessing rich prior knowledge obtained from pretraining with Internetscale corpus, have the potential to be a good controller with proper prompts. In this paper, we take HVAC Heating, Ventilation, and Air Conditioning building control as an example to examine the ability of GPT4 one of the firsttier foundation models as the controller. To control HVAC, we wrap the task as a language game by providing text including a short description for the task, several selected demonstrations, and the current observation to GPT4 on each step and execute the actions responded by GPT4. We conduct series of experiments to answer the following questions 1How well can GPT4 control HVAC 2How well can GPT4 generalize to different scenarios for HVAC control 3 How different parts of the text context affect the performance In general, we found GPT4 achieves the performance comparable to RL methods with few samples and low technical debt, indicating the potential of directly applying foundation models to industrial control tasks.

Large Language Model Prompt Chaining for Long Legal Document Classification ; Prompting is used to guide or steer a language model in generating an appropriate response that is consistent with the desired outcome. Chaining is a strategy used to decompose complex tasks into smaller, manageable components. In this study, we utilize prompt chaining for extensive legal document classification tasks, which present difficulties due to their intricate domainspecific language and considerable length. Our approach begins with the creation of a concise summary of the original document, followed by a semantic search for related exemplar texts and their corresponding annotations from a training corpus. Finally, we prompt for a label based on the task to assign, by leveraging the incontext learning from the fewshot prompt. We demonstrate that through prompt chaining, we can not only enhance the performance over zeroshot, but also surpass the microF1 score achieved by larger models, such as ChatGPT zeroshot, using smaller models.

VAST Vivify Your Talking Avatar via ZeroShot Expressive Facial Style Transfer ; Current talking face generation methods mainly focus on speechlip synchronization. However, insufficient investigation on the facial talking style leads to a lifeless and monotonous avatar. Most previous works fail to imitate expressive styles from arbitrary video prompts and ensure the authenticity of the generated video. This paper proposes an unsupervised variational style transfer model VAST to vivify the neutral photorealistic avatars. Our model consists of three key components a style encoder that extracts facial style representations from the given video prompts; a hybrid facial expression decoder to model accurate speechrelated movements; a variational style enhancer that enhances the style space to be highly expressive and meaningful. With our essential designs on facial style learning, our model is able to flexibly capture the expressive facial style from arbitrary video prompts and transfer it onto a personalized image renderer in a zeroshot manner. Experimental results demonstrate the proposed approach contributes to a more vivid talking avatar with higher authenticity and richer expressiveness.

Study of Jupiter's Interior with Quadratic Monte Carlo Simulations ; We construct models for Jupiter's interior that match the gravity data obtained by the Juno and Galileo spacecrafts. To generate ensembles of models, we introduce a novel quadratic Monte Carlo technique that is more efficient in confining fitness landscapes than affine invariant method that relies on linear stretch moves. We compare how long it takes the ensembles of walkers in both methods to travel to the most relevant parameter region. Once there, we compare the autocorrelation time and error bars of the two methods. For a ring potential and the 2d Rosenbrock function, we find that our quadratic Monte Carlo technique is significantly more efficient. Furthermore we modified the walk moves by adding a scaling factor. We provide the source code and examples so that this method can be applied elsewhere. Here we employ our method to generate fivelayer models for Jupiter's interior that include winds and a prominent dilute core, which allows us to match the planet's even and odd gravity harmonics. We compare predictions from the different model ensembles and analyze how much an increase of the temperature at 1 bar and ad hoc change to the equation of state affects the inferred amount of heavy elements in atmosphere and in the planet overall.

A study of dissipative models based on Dirac matrices ; We generalize the recent work of Shibata and Katsura, who considered a S12 chain with alternating XX and YY couplings in the presence of dephasing, the dynamics of which are described by the GKLS master equation. Their model is equivalent to a nonHermitian system described by the Kitaev formulation in terms of a single Majorana species hopping on a twoleg ladder in the presence of a nondynamical Z2 gauge field. Our generalization involves Dirac gamma matrix spin' operators on the square lattice, and maps onto a nonHermitian square lattice bilayer which is also Kitaevsolvable. We describe the exponentially many nonequilibrium steady states in this model. We identify how the spin degrees of freedom can be accounted for in the 2d model in terms of the gaugeinvariant quantities and then proceed to study the Liouvillian spectrum. We use a genetic algorithm to estimate the Liouvillian gap and the first decay modes for large system sizes. We observe a transition in the first decay modes, similar to that found by Shibata and Katsura. The results we obtain are consistent with a perturbative analysis for small and large values of the dissipation strength.

PlankAssembly Robust 3D Reconstruction from Three Orthographic Views with Learnt Shape Programs ; In this paper, we develop a new method to automatically convert 2D line drawings from three orthographic views into 3D CAD models. Existing methods for this problem reconstruct 3D models by backprojecting the 2D observations into 3D space while maintaining explicit correspondence between the input and output. Such methods are sensitive to errors and noises in the input, thus often fail in practice where the input drawings created by human designers are imperfect. To overcome this difficulty, we leverage the attention mechanism in a Transformerbased sequence generation model to learn flexible mappings between the input and output. Further, we design shape programs which are suitable for generating the objects of interest to boost the reconstruction accuracy and facilitate CAD modeling applications. Experiments on a new benchmark dataset show that our method significantly outperforms existing ones when the inputs are noisy or incomplete.

A Robust Planning Model for Offshore Microgrid Considering Tidal Power and Desalination ; Increasing attention has been paid to resources on islands, thus microgrids on islands need to be invested. Different from onshore microgrids, offshore microgrids OM are usually abundant in ocean renewable energy ORE, such as offshore wind, tidal power generation TPG, etc. Moreover, some special loads such as seawater desalination unit SDU should be included. In this sense, this paper proposes a planning method for OM to minimize the investment cost while the ORE's fluctuation could be accommodated with robustness. First, a deterministic planning model DPM is formulated for the OM with TPG and SDU. A robust planning model RPM is then developed considering the uncertainties from both TPG and load demand. The Columnandconstraint generation CCG algorithm is then employed to solve the RPM, producing planning results for the OM that is robust against the worst scenario. Results of the case studies show that the investment and operation decisions of the proposed model are robust, and TPG shows good complementarity with the other RESs.

GITMol A Multimodal Large Language Model for Molecular Science with Graph, Image, and Text ; Large language models have made significant strides in natural language processing, paving the way for innovative applications including molecular representation and generation. However, most existing singlemodality approaches cannot capture the abundant and complex information in molecular data. Here, we introduce GITMol, a multimodal large language model that integrates the structure Graph, Image, and Text information, including the Simplified Molecular Input Line Entry System SMILES and molecular captions. To facilitate the integration of multimodal molecular data, we propose GITFormer, a novel architecture capable of mapping all modalities into a unified latent space. Our study develops an innovative anytolanguage molecular translation strategy and achieves a 1015 improvement in molecular captioning, a 510 accuracy increase in property prediction, and a 20 boost in molecule generation validity compared to baseline or singlemodality models.

UniBrain Unify Image Reconstruction and Captioning All in One Diffusion Model from Human Brain Activity ; Image reconstruction and captioning from brain activity evoked by visual stimuli allow researchers to further understand the connection between the human brain and the visual perception system. While deep generative models have recently been employed in this field, reconstructing realistic captions and images with both lowlevel details and high semantic fidelity is still a challenging problem. In this work, we propose UniBrain Unify Image Reconstruction and Captioning All in One Diffusion Model from Human Brain Activity. For the first time, we unify image reconstruction and captioning from visualevoked functional magnetic resonance imaging fMRI through a latent diffusion model termed Versatile Diffusion. Specifically, we transform fMRI voxels into text and image latent for lowlevel information and guide the backward diffusion process through fMRIbased image and text conditions derived from CLIP to generate realistic captions and images. UniBrain outperforms current methods both qualitatively and quantitatively in terms of image reconstruction and reports image captioning results for the first time on the Natural Scenes Dataset NSD dataset. Moreover, the ablation experiments and functional regionofinterest ROI analysis further exhibit the superiority of UniBrain and provide comprehensive insight for visualevoked brain decoding.

Confidence Contours UncertaintyAware Annotation for Medical Semantic Segmentation ; Medical image segmentation modeling is a highstakes task where understanding of uncertainty is crucial for addressing visual ambiguity. Prior work has developed segmentation models utilizing probabilistic or generative mechanisms to infer uncertainty from labels where annotators draw a singular boundary. However, as these annotations cannot represent an individual annotator's uncertainty, models trained on them produce uncertainty maps that are difficult to interpret. We propose a novel segmentation representation, Confidence Contours, which uses high and lowconfidence contours'' to capture uncertainty directly, and develop a novel annotation system for collecting contours. We conduct an evaluation on the Lung Image Dataset Consortium LIDC and a synthetic dataset. From an annotation study with 30 participants, results show that Confidence Contours provide high representative capacity without considerably higher annotator effort. We also find that generalpurpose segmentation models can learn Confidence Contours at the same performance level as standard singular annotations. Finally, from interviews with 5 medical experts, we find that Confidence Contour maps are more interpretable than Bayesian maps due to representation of structural uncertainty.

Learning the hub graphical Lasso model with the structured sparsity via an efficient algorithm ; Graphical models have exhibited their performance in numerous tasks ranging from biological analysis to recommender systems. However, graphical models with hub nodes are computationally difficult to fit, particularly when the dimension of the data is large. To efficiently estimate the hub graphical models, we introduce a twophase algorithm. The proposed algorithm first generates a good initial point via a dual alternating direction method of multipliers ADMM, and then warm starts a semismooth Newton SSN based augmented Lagrangian method ALM to compute a solution that is accurate enough for practical tasks. The sparsity structure of the generalized Jacobian ensures that the algorithm can obtain a nice solution very efficiently. Comprehensive experiments on both synthetic data and real data show that it obviously outperforms the existing stateoftheart algorithms. In particular, in some high dimensional tasks, it can save more than 70 of the execution time, meanwhile still achieves a highquality estimation.

Edit TemporalConsistent Videos with Image Diffusion Model ; Largescale texttoimage T2I diffusion models have been extended for textguided video editing, yielding impressive zeroshot video editing performance. Nonetheless, the generated videos usually show spatial irregularities and temporal inconsistencies as the temporal characteristics of videos have not been faithfully modeled. In this paper, we propose an elegant yet effective TemporalConsistent Video Editing TCVE method, to mitigate the temporal inconsistency challenge for robust textguided video editing. In addition to the utilization of a pretrained 2D Unet for spatial content manipulation, we establish a dedicated temporal Unet architecture to faithfully capture the temporal coherence of the input video sequences. Furthermore, to establish coherence and interrelation between the spatialfocused and temporalfocused components, a cohesive joint spatialtemporal modeling unit is formulated. This unit effectively interconnects the temporal Unet with the pretrained 2D Unet, thereby enhancing the temporal consistency of the generated video output while simultaneously preserving the capacity for video content manipulation. Quantitative experimental results and visualization results demonstrate that TCVE achieves stateoftheart performance in both video temporal consistency and video editing capability, surpassing existing benchmarks in the field.

Diverse Cotraining Makes Strong SemiSupervised Segmentor ; Deep cotraining has been introduced to semisupervised segmentation and achieves impressive results, yet few studies have explored the working mechanism behind it. In this work, we revisit the core assumption that supports cotraining multiple compatible and conditionally independent views. By theoretically deriving the generalization upper bound, we prove the prediction similarity between two models negatively impacts the model's generalization ability. However, most current cotraining models are tightly coupled together and violate this assumption. Such coupling leads to the homogenization of networks and confirmation bias which consequently limits the performance. To this end, we explore different dimensions of cotraining and systematically increase the diversity from the aspects of input domains, different augmentations and model architectures to counteract homogenization. Our Diverse Cotraining outperforms the stateoftheart SOTA methods by a large margin across different evaluation protocols on the Pascal and Cityscapes. For example. we achieve the best mIoU of 76.2, 77.7 and 80.2 on Pascal with only 92, 183 and 366 labeled images, surpassing the previous best results by more than 5.

MeDM Mediating Image Diffusion Models for VideotoVideo Translation with Temporal Correspondence Guidance ; This study introduces an efficient and effective method, MeDM, that utilizes pretrained image Diffusion Models for videotovideo translation with consistent temporal flow. The proposed framework can render videos from scene position information, such as a normal Gbuffer, or perform textguided editing on videos captured in realworld scenarios. We employ explicit optical flows to construct a practical coding that enforces physical constraints on generated frames and mediates independent framewise scores. By leveraging this coding, maintaining temporal consistency in the generated videos can be framed as an optimization problem with a closedform solution. To ensure compatibility with Stable Diffusion, we also suggest a workaround for modifying observedspace scores in latentspace Diffusion Models. Notably, MeDM does not require finetuning or testtime optimization of the Diffusion Models. Through extensive qualitative, quantitative, and subjective experiments on various benchmarks, the study demonstrates the effectiveness and superiority of the proposed approach. Project page can be found at httpsmedm2023.github.io

Linking fast and slow the case for generative models ; A pervasive challenge in neuroscience is testing whether neuronal connectivity changes over time due to specific causes, such as stimuli, events, or clinical interventions. Recent hardware innovations and falling data storage costs enable longer, more naturalistic neuronal recordings. The implicit opportunity for understanding the selforganised brain calls for new analysis methods that link temporal scales from the order of milliseconds over which neuronal dynamics evolve, to the order of minutes, days or even years over which experimental observations unfold. This review article demonstrates how hierarchical generative models and Bayesian inference help to characterise neuronal activity across different time scales. Crucially, these methods go beyond describing statistical associations among observations and enable inference about underlying mechanisms. We offer an overview of fundamental concepts in statespace modeling and suggest a taxonomy for these methods. Additionally, we introduce key mathematical principles that underscore a separation of temporal scales, such as the slaving principle, and review Bayesian methods that are being used to test hypotheses about the brain with multiscale data. We hope that this review will serve as a useful primer for experimental and computational neuroscientists on the state of the art and current directions of travel in the complex systems modelling literature.

Can Language Models Learn to Listen ; We present a framework for generating appropriate facial responses from a listener in dyadic social interactions based on the speaker's words. Given an input transcription of the speaker's words with their timestamps, our approach autoregressively predicts a response of a listener a sequence of listener facial gestures, quantized using a VQVAE. Since gesture is a language component, we propose treating the quantized atomic motion elements as additional language token inputs to a transformerbased large language model. Initializing our transformer with the weights of a language model pretrained only on text results in significantly higher quality listener responses than training a transformer from scratch. We show that our generated listener motion is fluent and reflective of language semantics through quantitative metrics and a qualitative user study. In our evaluation, we analyze the model's ability to utilize temporal and semantic aspects of spoken text. Project page httpspeople.eecs.berkeley.eduevonnengprojectstext2listen

Building Emotional Support Chatbots in the Era of LLMs ; The integration of emotional support into various conversational scenarios presents profound societal benefits, such as social interactions, mental health counseling, and customer service. However, there are unsolved challenges that hinder realworld applications in this field, including limited data availability and the absence of wellaccepted model training paradigms. This work endeavors to navigate these challenges by harnessing the capabilities of Large Language Models LLMs. We introduce an innovative methodology that synthesizes human insights with the computational prowess of LLMs to curate an extensive emotional support dialogue dataset. Our approach is initiated with a meticulously designed set of dialogues spanning diverse scenarios as generative seeds. By utilizing the incontext learning potential of ChatGPT, we recursively generate an ExTensible Emotional Support dialogue dataset, named ExTES. Following this, we deploy advanced tuning techniques on the LLaMA model, examining the impact of diverse training strategies, ultimately yielding an LLM meticulously optimized for emotional support interactions. An exhaustive assessment of the resultant model showcases its proficiency in offering emotional support, marking a pivotal step in the realm of emotional support bots and paving the way for subsequent research and implementations.

The McDIPPER A novel saturationbased 31D initial state model for Heavy Ion Collisions ; We present a new 3D resolved model for the initial state of ultrarelativistic heavyion collisions, based on the kperpfactorized Color Glass Condensate hybrid approach. The McDIPPER framework responds to the need for a rapidityresolved initialstate Monte Carlo event generator which can deposit the relevant conserved charges energy, charge and baryon densities both in the midrapidity and forwardbackward regions of the collision. This eventbyevent generator computes the gluon and anti quark phasespace densities using the IPSat model, from where the relevant conserved charges can be computed directly. In the present work we have included the leading order contributions to the light flavor parton densities. As a feature, the model can be systematically improved in the future by adding nexttoleading order calculations in the CGC hybrid framework, and extended to lower energies by including subeikonal corrections the channels included. We present relevant observables, such as the eccentricities and flow decorrelation, as tests of this new approach.

Electronic Structure Prediction of Multimillion Atom Systems Through Uncertainty Quantification Enabled Transfer Learning ; The ground state electron density obtainable using KohnSham Density Functional Theory KSDFT simulations contains a wealth of material information, making its prediction via machine learning ML models attractive. However, the computational expense of KSDFT scales cubically with system size which tends to stymie training data generation, making it difficult to develop quantifiably accurate ML models that are applicable across many scales and system configurations. Here, we address this fundamental challenge by employing transfer learning to leverage the multiscale nature of the training data. Our ML models employ descriptors involving simple scalar products, comprehensively sample system configurations through thermalization, and quantify uncertainty in electron density predictions using Bayesian neural networks. We show that our models incur significantly lower data generation costs while allowing confident and when verifiable, accurate predictions for a wide variety of bulk systems well beyond training, including systems with defects, different alloy compositions, and at unprecedented, multimillionatom scales.

Breaking the Bank with ChatGPT FewShot Text Classification for Finance ; We propose the use of conversational GPT models for easy and quick fewshot text classification in the financial domain using the Banking77 dataset. Our approach involves incontext learning with GPT3.5 and GPT4, which minimizes the technical expertise required and eliminates the need for expensive GPU computing while yielding quick and accurate results. Additionally, we finetune other pretrained, masked language models with SetFit, a recent contrastive learning technique, to achieve stateoftheart results both in fulldata and fewshot settings. Our findings show that querying GPT3.5 and GPT4 can outperform finetuned, nongenerative models even with fewer examples. However, subscription fees associated with these solutions may be considered costly for small organizations. Lastly, we find that generative models perform better on the given task when shown representative samples selected by a human expert rather than when shown random ones. We conclude that a our proposed methods offer a practical solution for fewshot tasks in datasets with limited label availability, and b our stateoftheart results can inspire future work in the area.

On the implied volatility of European and Asian call options under the stochastic volatility Bachelier model ; In this paper we study the shorttime behavior of the atthemoney implied volatility for European and arithmetic Asian call options with fixed strike price. The asset price is assumed to follow the Bachelier model with a general stochastic volatility process. Using techniques of the Malliavin calculus such as the anticipating Ito's formula we first compute the level of the implied volatility when the maturity converges to zero. Then, we find a short maturity asymptotic formula for the skew of the implied volatility that depends on the roughness of the volatility model. We apply our general results to the SABR and fractional Bergomi models, and provide some numerical simulations that confirm the accurateness of the asymptotic formula for the skew.

Entanglement Verification with Deep Semisupervised Machine Learning ; Quantum entanglement lies at the heart in quantum information processing tasks. Although many criteria have been proposed, efficient and scalable methods to detect the entanglement of generally given quantum states are still not available yet, particularly for highdimensional and multipartite quantum systems. Based on FixMatch and PseudoLabel method, we propose a deep semisupervised learning model with a small portion of labeled data and a large portion of unlabeled data. The data augmentation strategies are applied in this model by using the convexity of separable states and performing local unitary operations on the training data. We verify that our model has good generalization ability and gives rise to better accuracies compared to traditional supervised learning models by detailed examples.

PAVI PlateAmortized Variational Inference ; Given observed data and a probabilistic generative model, Bayesian inference searches for the distribution of the model's parameters that could have yielded the data. Inference is challenging for large population studies where millions of measurements are performed over a cohort of hundreds of subjects, resulting in a massive parameter space. This large cardinality renders offtheshelf Variational Inference VI computationally impractical. In this work, we design structured VI families that efficiently tackle large population studies. Our main idea is to share the parameterization and learning across the different i.i.d. variables in a generative model, symbolized by the model's textitplates. We name this concept textitplate amortization. Contrary to offtheshelf stochastic VI, which slows down inference, plate amortization results in orders of magnitude faster to train variational distributions. Applied to largescale hierarchical problems, PAVI yields expressive, parsimoniously parameterized VI with an affordable training time. This faster convergence effectively unlocks inference in those large regimes. We illustrate the practical utility of PAVI through a challenging Neuroimaging example featuring 400 million latent parameters, demonstrating a significant step towards scalable and expressive Variational Inference.

TexttoOverpassQL A Natural Language Interface for Complex Geodata Querying of OpenStreetMap ; We present TexttoOverpassQL, a task designed to facilitate a natural language interface for querying geodata from OpenStreetMap OSM. The Overpass Query Language OverpassQL allows users to formulate complex database queries and is widely adopted in the OSM ecosystem. Generating Overpass queries from natural language input serves multiple usecases. It enables novice users to utilize OverpassQL without prior knowledge, assists experienced users with crafting advanced queries, and enables toolaugmented large language models to access information stored in the OSM database. In order to assess the performance of current sequence generation models on this task, we propose OverpassNL, a dataset of 8,352 queries with corresponding natural language inputs. We further introduce task specific evaluation metrics and ground the evaluation of the TexttoOverpassQL task by executing the queries against the OSM database. We establish strong baselines by finetuning sequencetosequence models and adapting large language models with incontext examples. The detailed evaluation reveals strengths and weaknesses of the considered learning strategies, laying the foundations for further research into the TexttoOverpassQL task.

VGDiffZero Texttoimage Diffusion Models Can Be Zeroshot Visual Grounders ; Largescale texttoimage diffusion models have shown impressive capabilities across various generative tasks, enabled by strong visionlanguage alignment obtained through pretraining. However, most visionlanguage discriminative tasks require extensive finetuning on carefullylabeled datasets to acquire such alignment, with great cost in time and computing resources. In this work, we explore directly applying a pretrained generative diffusion model to the challenging discriminative task of visual grounding without any finetuning and additional training dataset. Specifically, we propose VGDiffZero, a simple yet effective zeroshot visual grounding framework based on texttoimage diffusion models. We also design a comprehensive regionscoring method considering both global and local contexts of each isolated proposal. Extensive experiments on RefCOCO, RefCOCO, and RefCOCOg show that VGDiffZero achieves strong performance on zeroshot visual grounding.

A generalized vectorfield framework for mobility ; Trip flow between areas is a fundamental metric for human mobility research. Given its identification with travel demand and its relevance for transportation and urban planning, many models have been developed for its estimation. These models focus on flow intensity, disregarding the information provided by the local mobility orientation. A fieldtheoretic approach can overcome this issue and handling both intensity and direction at once. Here we propose a general vectorfield representation starting from individuals' trajectories valid for any type of mobility. By introducing four models of spatial exploration, we show how individuals' elections determine the mesoscopic properties of the mobility field. Distance optimization in long displacements and randomlike local exploration are necessary to reproduce empirical field features observed in Chinese logistic data and in New York City Foursquare checkins. Our framework is an essential tool to capture hidden symmetries in mesoscopic urban mobility, it establishes a benchmark to test the validity of mobility models and opens the doors to the use of field theory in a wide spectrum of applications.

Softmax Bias Correction for Quantized Generative Models ; Posttraining quantization PTQ is the goto compression technique for large generative models, such as stable diffusion or large language models. PTQ methods commonly keep the softmax activation in higher precision as it has been shown to be very sensitive to quantization noise. However, this can lead to a significant runtime and power overhead during inference on resourceconstraint edge devices. In this work, we investigate the source of the softmax sensitivity to quantization and show that the quantization operation leads to a large bias in the softmax output, causing accuracy degradation. To overcome this issue, we propose an offline bias correction technique that improves the quantizability of softmax without additional compute during deployment, as it can be readily absorbed into the quantization parameters. We demonstrate the effectiveness of our method on stable diffusion v1.5 and 125Msize OPT language model, achieving significant accuracy improvement for 8bit quantized softmax.

CIEM Contrastive Instruction Evaluation Method for Better Instruction Tuning ; Nowadays, the research on Large VisionLanguage Models LVLMs has been significantly promoted thanks to the success of Large Language Models LLM. Nevertheless, these VisionLanguage Models VLMs are suffering from the drawback of hallucination due to insufficient understanding of vision and language modalities, VLMs may generate incorrect perception information when doing downstream applications, for example, captioning a nonexistent entity. To address the hallucination phenomenon, on the one hand, we introduce a Contrastive Instruction Evaluation Method CIEM, which is an automatic pipeline that leverages an annotated imagetext dataset coupled with an LLM to generate factualcontrastive questionanswer pairs for the evaluation of the hallucination of VLMs. On the other hand, based on CIEM, we further propose a new instruction tuning method called CIT the abbreviation of Contrastive Instruction Tuning to alleviate the hallucination of VLMs by automatically producing highquality factualcontrastive questionanswer pairs and corresponding justifications for model tuning. Through extensive experiments on CIEM and CIT, we pinpoint the hallucination issues commonly present in existing VLMs, the disability of the current instructiontuning dataset to handle the hallucination phenomenon and the superiority of CITtuned VLMs over both CIEM and public datasets.

Breaking Barriers to Creative Expression CoDesigning and Implementing an Accessible TexttoImage Interface ; Texttoimage generation models have grown in popularity due to their ability to produce highquality images from a text prompt. One use for this technology is to enable the creation of more accessible art creation software. In this paper, we document the development of an alternative user interface that reduces the typing effort needed to enter image prompts by providing suggestions from a large language model, developed through iterative design and testing within the project team. The results of this testing demonstrate how generative text models can support the accessibility of texttoimage models, enabling users with a range of abilities to create visual art.

GRASS Unified Generation Model for SpeechtoSemantic Tasks ; This paper explores the instruction finetuning technique for speechtosemantic tasks by introducing a unified endtoend E2E framework that generates target text conditioned on a taskrelated prompt for audio data. We pretrain the model using large and diverse data, where instructionspeech pairs are constructed via a texttospeech TTS system. Extensive experiments demonstrate that our proposed model achieves stateoftheart SOTA results on many benchmarks covering speech named entity recognition, speech sentiment analysis, speech question answering, and more, after finetuning. Furthermore, the proposed model achieves competitive performance in zeroshot and fewshot scenarios. To facilitate future work on instruction finetuning for speechtosemantic tasks, we release our instruction dataset and code.

Mean field limits of particlebased stochastic reactiondriftdiffusion models ; We consider particlebased stochastic reactiondriftdiffusion models where particles move via diffusion and drift induced by one and twobody potential interactions. The dynamics of the particles are formulated as measurevalued stochastic processes MVSPs, which describe the evolution of the singular, stochastic concentration fields of each chemical species. The mean field large population limit of such models is derived and proven, giving coarsegrained deterministic partial integrodifferential equations PIDEs for the limiting deterministic concentration fields' dynamics. We generalize previous studies on the mean field limit of models involving only diffusive motion, with care to formulating the MVSP representation to ensure detailed balance of reversible reactions in the presence of potentials. Our work illustrates the more general set of limiting PIDEs that arise in the mean field limit, demonstrating that the limiting macroscopic reactive interaction terms for reversible reactions obtain additional nonlinear concentrationdependent coefficients compared to the purelydiffusive case.

Curve Your Attention MixedCurvature Transformers for Graph Representation Learning ; Realworld graphs naturally exhibit hierarchical or cyclical structures that are unfit for the typical Euclidean space. While there exist graph neural networks that leverage hyperbolic or spherical spaces to learn representations that embed such structures more accurately, these methods are confined under the messagepassing paradigm, making the models vulnerable against sideeffects such as oversmoothing and oversquashing. More recent work have proposed global attentionbased graph Transformers that can easily model longrange interactions, but their extensions towards nonEuclidean geometry are yet unexplored. To bridge this gap, we propose Fully ProductStereographic Transformer, a generalization of Transformers towards operating entirely on the product of constant curvature spaces. When combined with tokenized graph Transformers, our model can learn the curvature appropriate for the input graph in an endtoend fashion, without the need of additional tuning on different curvature initializations. We also provide a kernelized approach to nonEuclidean attention, which enables our model to run in time and memory cost linear to the number of nodes and edges while respecting the underlying geometry. Experiments on graph reconstruction and node classification demonstrate the benefits of generalizing Transformers to the nonEuclidean domain.

MaskDiffusion Boosting TexttoImage Consistency with Conditional Mask ; Recent advancements in diffusion models have showcased their impressive capacity to generate visually striking images. Nevertheless, ensuring a close match between the generated image and the given prompt remains a persistent challenge. In this work, we identify that a crucial factor leading to the textimage mismatch issue is the inadequate crossmodality relation learning between the prompt and the output image. To better align the prompt and image content, we advance the crossattention with an adaptive mask, which is conditioned on the attention maps and the prompt embeddings, to dynamically adjust the contribution of each text token to the image features. This mechanism explicitly diminishes the ambiguity in semantic information embedding from the text encoder, leading to a boost of texttoimage consistency in the synthesized images. Our method, termed MaskDiffusion, is trainingfree and hotpluggable for popular pretrained diffusion models. When applied to the latent diffusion models, our MaskDiffusion can significantly improve the texttoimage consistency with negligible computation overhead compared to the original diffusion models.

ZeroShot Cosalient Object Detection Framework ; Cosalient Object Detection CoSOD endeavors to replicate the human visual system's capacity to recognize common and salient objects within a collection of images. Despite recent advancements in deep learning models, these models still rely on training with wellannotated CoSOD datasets. The exploration of trainingfree zeroshot CoSOD frameworks has been limited. In this paper, taking inspiration from the zeroshot transfer capabilities of foundational computer vision models, we introduce the first zeroshot CoSOD framework that harnesses these models without any training process. To achieve this, we introduce two novel components in our proposed framework the group prompt generation GPG module and the cosaliency map generation CMP module. We evaluate the framework's performance on widelyused datasets and observe impressive results. Our approach surpasses existing unsupervised methods and even outperforms fully supervised methods developed before 2020, while remaining competitive with some fully supervised methods developed before 2022.

A Minimal Model for Understanding Secondary Cosmic Rays ; We take a phenomenological approach in a minimal model to understand the spectral intensity of secondary cosmicray particles like positrons, antiprotons, Lithium, Beryllium and Boron. Our analysis shows that cosmic rays at sim GeV energies pass through a significant amount of matter in regions surrounding the sources. This grammage decreases with increasing cosmicray energy and becomes negligible beyond sim 100 GeV. During the subsequent propagation in the interstellar medium cosmic rays of all energies up to sim 105 GeVn pass through about 12 g cm2 of matter before leaking into the intergalactic medium. It is in the interstellar medium that the bulk of the positrons and antiprotons are generated. Also cosmicray nuclei like C, N, and O at all energies generate additional amounts of Li, Be and B nuclei with a spectrum similar to those of C, O etc. The implications of these findings of the minimal model to the observations of gamma rays and also the importance of spatial and temporal discreteness of cosmicray sources for modeling cosmicray propagation are briefly pointed out.

Antiglitch a Quasiphysical Model for Removing Short Glitches from LIGO and Virgo Data ; Gravitationalwave observatories become more sensitive with each observing run, increasing the number of detected gravitationalwave signals. A limiting factor in identifying these signals is the presence of transient nonGaussian noise, which generates glitches that can mimic gravitational wave signals. Our work provides a quasiphysical model waveform for the four most common types of short transient glitches, which are particularly problematic in the search for highmass black hole binaries. Our model has only a few, physically interpretable parameters central frequency, bandwidth, phase, amplitude and time. We demonstrate the accuracy of this model by fitting and removing a large sample of glitches from a month of LIGO and Virgo data from the O3 observing run. We can effectively remove three of the four types of short transients. We finally map the ability of these glitches to mimic binary black hole signals.

Semantic Adversarial Attacks via Diffusion Models ; Traditional adversarial attacks concentrate on manipulating clean examples in the pixel space by adding adversarial perturbations. By contrast, semantic adversarial attacks focus on changing semantic attributes of clean examples, such as color, context, and features, which are more feasible in the real world. In this paper, we propose a framework to quickly generate a semantic adversarial attack by leveraging recent diffusion models since semantic information is included in the latent space of welltrained diffusion models. Then there are two variants of this framework 1 the Semantic Transformation ST approach finetunes the latent space of the generated image andor the diffusion model itself; 2 the Latent Masking LM approach masks the latent space with another target image and local backpropagationbased interpretation methods. Additionally, the ST approach can be applied in either whitebox or blackbox settings. Extensive experiments are conducted on CelebAHQ and AFHQ datasets, and our framework demonstrates great fidelity, generalizability, and transferability compared to other baselines. Our approaches achieve approximately 100 attack success rate in multiple settings with the best FID as 36.61. Code is available at httpsgithub.comsteven202semanticadvviadm.

Voxtlm unified decoderonly models for consolidating speech recognitionsynthesis and speechtext continuation tasks ; We propose a decoderonly language model, textitVoxtLM, that can perform four tasks speech recognition, speech synthesis, text generation, and speech continuation. VoxtLM integrates text vocabulary with discrete speech tokens from selfsupervised speech features and uses special tokens to enable multitask learning. Compared to a singletask model, VoxtLM exhibits a significant improvement in speech synthesis, with improvements in both speech intelligibility from 28.9 to 5.6 and objective quality from 2.68 to 3.90. VoxtLM also improves speech generation and speech recognition performance over the singletask counterpart. VoxtLM is trained with publicly available data and training recipes and model checkpoints will be opensourced to make fully reproducible work.

Common origin of dark matter, baryon asymmetry and neutrino masses in the standard model with extended scalars ; We propose a model that simultaneously addresses the existence of a dark matter candidate, baryon asymmetry and tiny neutrino masses and mixing by introducing two SU2 triplet scalars and an inert SU2 doublet scalar on top of the standard model. The two triplet scalars serve as mediators in generation of lepton asymmetry and determination of relic density of dark matter. They also play an essential role in generation of tiny neutrino masses and inducing CP violation. The inert scalar is regarded as a dark matter candidate. The interference due to complex BreitWigner propagators for the triplets will result in CPasymmetry that depends on the difference between their masses and a relative complex phase between their couplings to standard model leptons. Moreover, the production of lepton asymmetry will be closely tied to the evolution of dark matter, limiting the parameter space where the correct relic abundance and matterantimatter asymmetry can be simultaneously accomplished.

Toward responsible face datasets modeling the distribution of a disentangled latent space for sampling face images from demographic groups ; Recently, it has been exposed that some modern facial recognition systems could discriminate specific demographic groups and may lead to unfair attention with respect to various facial attributes such as gender and origin. The main reason are the biases inside datasets, unbalanced demographics, used to train theses models. Unfortunately, collecting a largescale balanced dataset with respect to various demographics is impracticable. In this paper, we investigate as an alternative the generation of a balanced and possibly biasfree synthetic dataset that could be used to train, to regularize or to evaluate deep learningbased facial recognition models. We propose to use a simple method for modeling and sampling a disentangled projection of a StyleGAN latent space to generate any combination of demographic groups e.g. hispanicfemale. Our experiments show that we can synthesis any combination of demographic groups effectively and the identities are different from the original training dataset. We also released the source code.

MusiLingo Bridging Music and Text with Pretrained Language Models for Music Captioning and Query Response ; Large Language Models LLMs have shown immense potential in multimodal applications, yet the convergence of textual and musical domains remains relatively unexplored. To address this gap, we present MusiLingo, a novel system for music caption generation and musicrelated query responses. MusiLingo employs a single projection layer to align music representations from the pretrained frozen music audio model MERT with the frozen LLaMA language model, bridging the gap between music audio and textual contexts. We train it on an extensive music caption dataset and finetune it with instructional data. Due to the scarcity of highquality music QA datasets, we created the MusicInstruct MI dataset from MusicCaps, tailored for openended music inquiries. Empirical evaluations demonstrate its competitive performance in generating music captions and composing musicrelated QA pairs. Our introduced dataset enables notable advancements beyond previous ones.

Exploring the impact of lowrank adaptation on the performance, efficiency, and regularization of RLHF ; During the last stage of RLHF, a large language model is aligned to human intents via PPO training, a process that generally requires largescale computational resources. In this technical report, we empirically investigate an efficient implementation of RLHF using lowrank adaptation LoRA, which allows us to align the LLaMA 7B checkpoint on the Alpaca dataset using only two A100 GPUs instead of the eight required for full model finetuning. Despite tuning only 0.2 of LLaMA 7B's parameters, our implementation achieves better performance than the publiclyreleased AlpacaFarm checkpoint with full model finetuning. Next, we analyze several configurations of our LoRAbased PPO implementation, varying the form of the KL regularization term in the training objective. We find that 1 removing this penalty term does not harm performance on the AlpacaFarm evaluation set under our LoRA setup; 2 other regularizers, such as JensenShannon divergence, lead to improved performance; and 3 while PPO training negatively impacts the factuality of modelgenerated responses, training with LoRA largely mitigates this effect. We release our code and pretrained checkpoints to facilitate future research on more efficient RLHF.

Mitigating Shortcuts in Language Models with Soft Label Encoding ; Recent research has shown that large language models rely on spurious correlations in the data for natural language understanding NLU tasks. In this work, we aim to answer the following research question Can we reduce spurious correlations by modifying the ground truth labels of the training data Specifically, we propose a simple yet effective debiasing framework, named Soft Label Encoding SoftLE. We first train a teacher model with hard labels to determine each sample's degree of relying on shortcuts. We then add one dummy class to encode the shortcut degree, which is used to smooth other dimensions in the ground truth label to generate soft labels. This new ground truth label is used to train a more robust student model. Extensive experiments on two NLU benchmark tasks demonstrate that SoftLE significantly improves outofdistribution generalization while maintaining satisfactory indistribution accuracy.

ModelBased Generation of AttackFault Trees ; Joint safety and security analysis of cyberphysical systems is a necessary step to correctly capture interdependencies between these properties. AttackFault Trees represent a combination of dynamic Fault Trees and Attack Trees and can be used to model and modelcheck a holistic view on both safety and security. Manually creating a complete AFT for the whole system is, however, a daunting task. It needs to span multiple abstraction layers, e.g., abstract application architecture and data flow as well as system and library dependencies that are affected by various vulnerabilities. We present an AFT generation toolchain that facilitates this task using partial Fault and Attack Trees that are either manually created or mined from vulnerability databases. We semiautomatically create two system models that provide the necessary information to automatically combine these partial Fault and Attack Trees into complete AFTs using graph transformation rules.

Prompt a Robot to Walk with Large Language Models ; Large language models LLMs pretrained on vast internetscale data have showcased remarkable capabilities across diverse domains. Recently, there has been escalating interest in deploying LLMs for robotics, aiming to harness the power of foundation models in realworld settings. However, this approach faces significant challenges, particularly in grounding these models in the physical world and in generating dynamic robot motions. To address these issues, we introduce a novel paradigm in which we use fewshot prompts collected from the physical environment, enabling the LLM to autoregressively generate lowlevel control commands for robots without taskspecific finetuning. Experiments across various robots and environments validate that our method can effectively prompt a robot to walk. We thus illustrate how LLMs can proficiently function as lowlevel feedback controllers for dynamic motion control even in highdimensional robotic systems. The project website and source code can be found at httpsprompt2walk.github.io .

Generative AI in Mafialike Game Simulation ; In this research, we explore the efficacy and potential of Generative AI models, specifically focusing on their application in roleplaying simulations exemplified through Spyfall, a renowned mafiastyle game. By leveraging GPT4's advanced capabilities, the study aimed to showcase the model's potential in understanding, decisionmaking, and interaction during game scenarios. Comparative analyses between GPT4 and its predecessor, GPT3.5turbo, demonstrated GPT4's enhanced adaptability to the game environment, with significant improvements in posing relevant questions and forming humanlike responses. However, challenges such as the model;s limitations in bluffing and predicting opponent moves emerged. Reflections on game development, financial constraints, and nonverbal limitations of the study were also discussed. The findings suggest that while GPT4 exhibits promising advancements over earlier models, there remains potential for further development, especially in instilling more humanlike attributes in AI.

Stochastic compressible NavierStokes equations under location uncertainty and its approximations for ocean modelling ; The aim of this paper is to provide a stochastic version under location uncertainty of the compressible NavierStokes equations. The modelling under location uncertainty setting is used here to derive a physically consistent stochastic dynamical system for compressible flows. It relies on an extended stochastic version of the Reynolds transport theorem involving stochastic source terms. In a similar way as in the deterministic case, this conservation theorem is applied to density, momentum and total energy in order to obtain a transport equation of the primitive variables, i.e. density, velocity and temperature. For the modelling of ocean dynamics, the transport of mass fraction of species, such as salinity, is also considered. We show that performing low Mach and Boussinesq approximations to this more general set of equations allows us to recover previous versions of incompressible stochastic NavierStokes equations and the stochastic Boussinesq equations, respectively. Finally, we provide some research directions on the use of this general set of equations in the perspective of relaxing the Boussinesq and hydrostatic approximation for ocean modelling.

Scaling Limits of the Wasserstein information matrix on Gaussian Mixture Models ; We consider the Wasserstein metric on the Gaussian mixture models GMMs, which is defined as the pullback of the full Wasserstein metric on the space of smooth probability distributions with finite second moment. It derives a class of Wasserstein metrics on probability simplices over onedimensional bounded homogeneous lattices via a scaling limit of the Wasserstein metric on GMMs. Specifically, for a sequence of GMMs whose variances tend to zero, we prove that the limit of the Wasserstein metric exists after certain renormalization. Generalizations of this metric in general GMMs are established, including inhomogeneous lattice models whose lattice gaps are not the same, extended GMMs whose mean parameters of Gaussian components can also change, and the secondorder metric containing highorder information of the scaling limit. We further study the Wasserstein gradient flows on GMMs for three typical functionals potential, internal, and interaction energies. Numerical examples demonstrate the effectiveness of the proposed GMM models for approximating Wasserstein gradient flows.

Deep3DSketch Rapid 3D Modeling from Single Freehand Sketches ; The rapid development of ARVR brings tremendous demands for 3D content. While the widelyused ComputerAided Design CAD method requires a timeconsuming and laborintensive modeling process, sketchbased 3D modeling offers a potential solution as a natural form of computerhuman interaction. However, the sparsity and ambiguity of sketches make it challenging to generate highfidelity content reflecting creators' ideas. Precise drawing from multiple views or strategic stepbystep drawings is often required to tackle the challenge but is not friendly to novice users. In this work, we introduce a novel endtoend approach, Deep3DSketch, which performs 3D modeling using only a single freehand sketch without inputting multiple sketches or view information. Specifically, we introduce a lightweight generation network for efficient inference in realtime and a structuralaware adversarial training approach with a Stroke Enhancement Module SEM to capture the structural information to facilitate learning of the realistic and finedetailed shape structures for highfidelity performance. Extensive experiments demonstrated the effectiveness of our approach with the stateoftheart SOTA performance on both synthetic and real datasets.

Memoryaugmented conformer for improved endtoend longform ASR ; Conformers have recently been proposed as a promising modelling approach for automatic speech recognition ASR, outperforming recurrent neural networkbased approaches and transformers. Nevertheless, in general, the performance of these endtoend models, especially attentionbased models, is particularly degraded in the case of long utterances. To address this limitation, we propose adding a fullydifferentiable memoryaugmented neural network between the encoder and decoder of a conformer. This external memory can enrich the generalization for longer utterances since it allows the system to store and retrieve more information recurrently. Notably, we explore the neural Turing machine NTM that results in our proposed ConformerNTM model architecture for ASR. Experimental results using Librispeech trainclean100 and train960 sets show that the proposed system outperforms the baseline conformer without memory for long utterances.

AntiIto noiseinduced phase transitions in tumor growth with chemotherapy ; The objective of this work is to apply the HanggiKlimontovich stochastic differential equations to model and study the effects of antitumor chemotherapy in the case of continuous infusion delivering. The fluctuations generated by variations in drug concentration are modeled by the HanggiKlimontovich stochastic integral. This integral, which in the physics literature is sometimes called antiIto integral, in the last decade it has been referenced quite as the more appropriate stochastic integral for model various biological and physical systems. Then, we make some comparisons with the model based on Ito stochastic differential equations and their phase transitions that they generate, showing that the HanggiKlimontovich stochastic differential equations lead to more biologically realistic results.

COCOCounterfactuals Automatically Constructed Counterfactual Examples for ImageText Pairs ; Counterfactual examples have proven to be valuable in the field of natural language processing NLP for both evaluating and improving the robustness of language models to spurious correlations in datasets. Despite their demonstrated utility for NLP, multimodal counterfactual examples have been relatively unexplored due to the difficulty of creating paired imagetext data with minimal counterfactual changes. To address this challenge, we introduce a scalable framework for automatic generation of counterfactual examples using texttoimage diffusion models. We use our framework to create COCOCounterfactuals, a multimodal counterfactual dataset of paired image and text captions based on the MSCOCO dataset. We validate the quality of COCOCounterfactuals through human evaluations and show that existing multimodal models are challenged by our counterfactual imagetext pairs. Additionally, we demonstrate the usefulness of COCOCounterfactuals for improving outofdomain generalization of multimodal visionlanguage models via training data augmentation.

Approximation Rates for Deep Calibration of Rough Stochastic Volatility Models ; We derive quantitative error bounds for deep neural networks DNNs approximating option prices on a ddimensional risky asset as functions of the underlying model parameters, payoff parameters and initial conditions. We cover a general class of stochastic volatility models of Markovian nature as well as the rough Bergomi model. In particular, under suitable assumptions we show that option prices can be learned by DNNs up to an arbitrary small error varepsilon in 0,12 while the network size grows only subpolynomially in the asset vector dimension d and the reciprocal varepsilon1 of the accuracy. Hence, the approximation does not suffer from the curse of dimensionality. As quantitative approximation results for DNNs applicable in our setting are formulated for functions on compact domains, we first consider the case of the asset price restricted to a compact set, then we extend these results to the general case by using convergence arguments for the option prices.

Scaling solutions as Early Dark Energy resolutions to the Hubble tension ; A wide class of scalar field models including Quintessence and Kessence have the attractive property of tracker regimes, where the energy density stored in the field evolves so as to mimic that of the dominant background component for a period of time. During this evolution, for a brief period of time there is an increase in the energy density of the field as it spirals in towards it's attractor solution. We show that when the peak of this energy density occurs around the epoch of equality, we can address a key requirement of early dark energy EDE, postulated as a solution to the Hubble tension. In particular we demonstrate how this can occur in a wide class of Quintessence, axion and Kessence models, before showing that the Quintessence models suffer in that they generally lead to sound speeds incompatible with the requirements of EDE, whereas the Kessence and axion models can do a better job of fitting the data.

Toward Robust Recommendation via Realtime Vicinal Defense ; Recommender systems have been shown to be vulnerable to poisoning attacks, where malicious data is injected into the dataset to cause the recommender system to provide biased recommendations. To defend against such attacks, various robust learning methods have been proposed. However, most methods are modelspecific or attackspecific, making them lack generality, while other methods, such as adversarial training, are oriented towards evasion attacks and thus have a weak defense strength in poisoning attacks. In this paper, we propose a general method, Realtime Vicinal Defense RVD, which leverages neighboring training data to finetune the model before making a recommendation for each user. RVD works in the inference phase to ensure the robustness of the specific sample in realtime, so there is no need to change the model structure and training process, making it more practical. Extensive experimental results demonstrate that RVD effectively mitigates targeted poisoning attacks across various models without sacrificing accuracy. Moreover, the defensive effect can be further amplified when our method is combined with other strategies.

AVCPL Continuous PseudoLabeling for AudioVisual Speech Recognition ; Audiovisual speech contains synchronized audio and visual information that provides crossmodal supervision to learn representations for both automatic speech recognition ASR and visual speech recognition VSR. We introduce continuous pseudolabeling for audiovisual speech recognition AVCPL, a semisupervised method to train an audiovisual speech recognition AVSR model on a combination of labeled and unlabeled videos with continuously regenerated pseudolabels. Our models are trained for speech recognition from audiovisual inputs and can perform speech recognition using both audio and visual modalities, or only one modality. Our method uses the same audiovisual model for both supervised training and pseudolabel generation, mitigating the need for external speech recognition models to generate pseudolabels. AVCPL obtains significant improvements in VSR performance on the LRS3 dataset while maintaining practical ASR and AVSR performance. Finally, using visualonly speech data, our method is able to leverage unlabeled visual speech to improve VSR.

Seismogram Transformer A generic deep learning backbone network for multiple earthquake monitoring tasks ; Seismic records, known as seismograms, are crucial records of ground motion resulting from seismic events, constituting the backbone of earthquake research and monitoring. The latest advancements in deep learning have significantly facilitated various seismic signal processing tasks. This paper introduces a novel backbone neural network model designed for various seismic monitoring tasks, named Seismogram Transformer SeisT. Thanks to its efficient network architecture, SeisT matches or even outperforms the stateoftheart models in earthquake detection, seismic phase picking, firstmotion polarity classification, magnitude estimation, and azimuth estimation tasks, particularly in terms of outofdistribution generalization performance. SeisT consists of multiple network layers composed of different foundational blocks, which help the model understand multilevel feature representations of seismograms from lowlevel to highlevel complex features, effectively extracting features such as frequency, phase, and timefrequency relationships from input seismograms. Three differentsized models were customized based on these diverse foundational modules. Through extensive experiments and performance evaluations, this study showcases the capabilities and potential of SeisT in advancing seismic signal processing and earthquake research.

Simulationbased Inference with the Generalized KullbackLeibler Divergence ; In Simulationbased Inference, the goal is to solve the inverse problem when the likelihood is only known implicitly. Neural Posterior Estimation commonly fits a normalized density estimator as a surrogate model for the posterior. This formulation cannot easily fit unnormalized surrogates because it optimizes the KullbackLeibler divergence. We propose to optimize a generalized KullbackLeibler divergence that accounts for the normalization constant in unnormalized distributions. The objective recovers Neural Posterior Estimation when the model class is normalized and unifies it with Neural Ratio Estimation, combining both into a single objective. We investigate a hybrid model that offers the best of both worlds by learning a normalized base distribution and a learned ratio. We also present benchmark results.