NeurIPS 2023 Papers — Page 32
Conference on Neural Information Processing Systems · 3218 papers
Unpaired Multi-Domain Causal Representation Learning
Nils Sturma (Technical University of Munich), Caroline Uhler (Massachusetts Institute of Technology)
Representation LearningContrastive LearningTabular
🎯 What it does: A framework for unpaired multi-domain causal representation learning is proposed, providing sufficient conditions for the identifiability of shared latent variables and causal graphs, and implementing a two-step algorithm based on linear ICA to recover shared latent variables and causal structures.
UNSSOR: Unsupervised Neural Speech Separation by Leveraging Over-determined Training Mixtures
Zhong-Qiu Wang (Carnegie Mellon University), Shinji Watanabe (Carnegie Mellon University)
RecognitionRestorationConvolutional Neural NetworkAudio
🎯 What it does: This paper proposes the UNSSOR method, which utilizes overdetermined mixed signals (more microphones than speakers) to train a neural network under unsupervised conditions by constructing a mixed constraint loss, achieving speech separation; after training, it can be used for single-microphone (underdetermined) separation.
Unsupervised Anomaly Detection with Rejection
Lorenzo Perini (KU Leuven), Jesse Davis (KU Leuven)
Anomaly DetectionImageTabularTime SeriesBiomedical DataFinance RelatedAudio
🎯 What it does: This paper proposes an unsupervised learning to reject framework REJEX, which sets a constant threshold in anomaly detection using the EXCEED stability measure, thereby enabling the rejection of predictions for highly uncertain samples.
Unsupervised Behavior Extraction via Random Intent Priors
Hao Hu (Tsinghua University), Chongjie Zhang (Washington University in St. Louis)
Robotic IntelligenceReinforcement LearningTabularBenchmark
🎯 What it does: The UBER method is proposed, which utilizes a random intent prior to generate diverse rewards from offline, reward-free data, extracts various behaviors using offline RL, and accelerates task learning in the online phase through policy reuse.
Unsupervised Graph Neural Architecture Search with Disentangled Self-Supervision
Zeyang Zhang (Tsinghua University), Wenwu Zhu (Tsinghua University)
Neural Architecture SearchGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A completely unsupervised graph neural network architecture search framework, DSGAS, is proposed, which can automatically discover the best GNN architecture suitable for different graph data without labels.
Unsupervised Image Denoising with Score Function
Yutong Xie (Peking University), Quanzheng Li (Massachusetts General Hospital and Harvard Medical School)
RestorationScore-based ModelAuto EncoderImage
🎯 What it does: A score function-based unsupervised image denoising method is proposed, which can obtain clean images by solving equations related to the noise model;
Unsupervised Learning for Solving the Travelling Salesman Problem
Yimeng Min (Cornell University), Carla P Gomes
OptimizationGraph Neural NetworkGraph
🎯 What it does: Train an unsupervised graph neural network to generate a heatmap representing the probability that each edge belongs to the optimal path, and further optimize it using local search to obtain an approximate optimal solution for the traveling salesman problem.
Unsupervised Optical Flow Estimation with Dynamic Timing Representation for Spike Camera
Lujie Xia (Peking University), Ruiqin Xiong (Peking University)
Autonomous DrivingConvolutional Neural NetworkSpiking Neural NetworkOptical FlowImageVideo
🎯 What it does: A novel unsupervised optical flow estimation method for event cameras, USFlow, is proposed, which utilizes multi-layer dilated convolutions to achieve dynamic temporal representation, extracts features from multi-scale temporal windows, and constructs a self-supervised loss through optical flow-guided light intensity fusion.
Unsupervised Polychromatic Neural Representation for CT Metal Artifact Reduction
Qing Wu (ShanghaiTech University), Yuyao Zhang (ShanghaiTech University)
RestorationImageComputed Tomography
🎯 What it does: Proposes an unsupervised multi-energy neural representation (Polyner) that directly recovers metal artifact-free images from CT projections affected by metal;
Unsupervised Protein-Ligand Binding Energy Prediction via Neural Euler's Rotation Equation
Wengong Jin (Broad Institute of MIT and Harvard), Caroline Uhler (Broad Institute of MIT and Harvard)
Drug DiscoveryProtein Structure PredictionGraph Neural NetworkBiomedical DataStochastic Differential Equation
🎯 What it does: An unsupervised framework for predicting protein-ligand (or antibody-antigen) binding energy is proposed, treating binding energy as the likelihood of the structure and using an energy-based model for training.
Unsupervised Semantic Correspondence Using Stable Diffusion
Eric Hedlin (University of British Columbia), Kwang Moo Yi (University of British Columbia)
GenerationRetrievalTransformerPrompt EngineeringDiffusion modelImage
🎯 What it does: Utilize attention maps from Stable Diffusion for unsupervised semantic correspondence matching;
Unsupervised Video Domain Adaptation for Action Recognition: A Disentanglement Perspective
Pengfei Wei (ByteDance), Xiang Yin (ByteDance)
RecognitionDomain AdaptationAuto EncoderContrastive LearningVideoMultimodality
🎯 What it does: This paper proposes an unsupervised video domain adaptation framework based on a separable variational autoencoder (TranSVAE), specifically for action recognition tasks.
UP-DP: Unsupervised Prompt Learning for Data Pre-Selection with Vision-Language Models
Xin Li (Bosch Research North America), Liu Ren (Bosch Research North America)
ClassificationRecognitionRetrievalTransformerPrompt EngineeringVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: This paper proposes an unsupervised prompt learning method (UP-DP) that learns trainable text prompts within the BLIP-2 vision-language model to generate better multimodal features, enabling efficient one-time data pre-selection on unlabeled data.
UP-NeRF: Unconstrained Pose Prior-Free Neural Radiance Field
Injae Kim (Korea University), Hyunwoo J. Kim (Korea University)
GenerationPose EstimationTransformerNeural Radiance FieldImage
🎯 What it does: A method called UP-NeRF is proposed, which jointly optimizes camera poses and neural radiance fields to achieve high-quality view synthesis without camera pose priors and in the presence of inconsistent lighting and transient occlusions.
Use perturbations when learning from explanations
Juyeon Heo (University of Cambridge), Adrian Weller (Alan Turing Institute)
Explainability and InterpretabilityImage
🎯 What it does: This paper rephrases the framework of Model Explanation (MLX) as a robustness problem, utilizing human-provided explanation masks to define the perturbation space, thereby training models that are robust to irrelevant features without requiring strong parameter smoothing.
User-Level Differential Privacy With Few Examples Per User
Badih Ghazi (Google Research), Chiyuan Zhang (Google Research)
OptimizationSafty and Privacy
🎯 What it does: This paper proposes a general transformation from item-level differential privacy (item-level DP) algorithms to user-level differential privacy (user-level DP) algorithms, focusing on scenarios with few user samples, and provides new theoretical analysis and upper bounds.
Using Imperfect Surrogates for Downstream Inference: Design-based Supervised Learning for Social Science Applications of Large Language Models
Naoki Egami (Columbia University), Hanying Wei (Columbia University)
Large Language ModelSupervised Fine-TuningText
🎯 What it does: A new algorithm is proposed that utilizes imperfect annotation substitutes for downstream statistical analysis while ensuring statistical properties such as asymptotic unbiasedness and appropriate uncertainty quantification, which are crucial for computational social science research.
Utilitarian Algorithm Configuration
Devon R. Graham (University of British Columbia), Tim Roughgarden (Columbia University)
OptimizationTabular
🎯 What it does: This paper proposes a utility function-based algorithm configuration method called Utilitarian Procrastination, addressing the shortcomings of traditional expected runtime minimization and providing theoretical guarantees.
V-InFoR: A Robust Graph Neural Networks Explainer for Structurally Corrupted Graphs
Senzhang Wang (Central South University), Jianxin Wang (Central South University)
Explainability and InterpretabilityAdversarial AttackGraph Neural NetworkAuto EncoderGraph
🎯 What it does: A robust GNN interpreter V-InfoR for structurally damaged graphs is proposed, which can still provide reliable explanatory subgraphs when the graph structure is compromised by noise or attacks.
VanillaNet: the Power of Minimalism in Deep Learning
Hanting Chen (Huawei Noah's Ark Lab), Dacheng Tao (University of Sydney)
Object DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: A minimalist convolutional network called VanillaNet is proposed, which removes depth, shortcut connections, and self-attention, using only a minimalist module of 1×1 convolution + BN + activation. A deep training strategy and a series of activation functions are employed during training to enhance non-linearity.
Variance-Reduced Gradient Estimation via Noise-Reuse in Online Evolution Strategies
Oscar Li (Carnegie Mellon University), Luke Metz (OpenAI)
OptimizationReinforcement LearningTime SeriesSequential
🎯 What it does: This paper proposes an online unbiased, low-variance evolutionary strategy gradient estimation method called NRES, which addresses the high variance and slow convergence issues of traditional online ES.
Variational Annealing on Graphs for Combinatorial Optimization
Sebastian Sanokowski (Johannes Kepler University Linz), Sebastian Lehner (Johannes Kepler University Linz)
OptimizationGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: The VAG-CO method is proposed, which solves combinatorial optimization problems using a self-regressive variational adaptive graph model.
Variational Gaussian processes for linear inverse problems
Thibault Christophe RANDRIANARISOA (Bocconi University), Botond Szabo (Bocconi University)
RestorationOptimizationTabular
🎯 What it does: This paper extends the variational Bayesian (VB) method based on inducing variables to linear inverse problems, theoretically derives the shrinkage rate of the variational posterior, and validates the optimal convergence properties in typical inverse problems such as the Poisson equation, heat equation, and Radon transform.
Variational Gaussian Processes with Decoupled Conditionals
Xinran Zhu (Cornell University), David Bindel (Cornell University)
OptimizationTabular
🎯 What it does: This paper proposes the use of decoupled conditionals in Variational Gaussian Processes (VGP), allowing the predictive mean and covariance to be learned independently, thereby enhancing the model's expressiveness.
Variational Imbalanced Regression: Fair Uncertainty Quantification via Probabilistic Smoothing
Ziyan Wang (Georgia Institute of Technology), Hao Wang (Rutgers University)
Auto EncoderImageText
🎯 What it does: The Variational Imbalanced Regression (VIR) model is proposed to address regression tasks with imbalanced label distributions, naturally producing reasonable uncertainty estimates in the process.
Variational Inference with Gaussian Score Matching
Chirag Modi (Flatiron Institute), Lawrence K. Saul (Flatiron Institute)
OptimizationScore-based ModelTabular
🎯 What it does: A variational inference method based on score matching (GSM-VI) is proposed, which achieves closed-form iterative updates of Gaussian family variational distributions by minimizing the KL distance at each step and enforcing the matching of the posterior and the gradient of the variational distribution.
Variational Monte Carlo on a Budget — Fine-tuning pre-trained Neural Wavefunctions
Michael Scherbela (University of Vienna), Philipp Grohs (University of Vienna)
Graph Neural NetworkSupervised Fine-TuningGraphPhysics Related
🎯 What it does: A deep learning-based variable component quantum Monte Carlo (DL-VMC) wave function model was trained, achieving zero-shot and few-shot energy predictions on new molecules.
Variational Weighting for Kernel Density Ratios
Sangwoong Yoon (Korea Institute for Advanced Study), Yung-Kyun Noh (Korea Institute for Advanced Study)
Object DetectionAnomaly DetectionOptimizationGenerative Adversarial NetworkImageTabular
🎯 What it does: This paper proposes Variational Weighted Kernel Density Estimation (VWKDE), which reduces bias in density ratio estimation by applying position-dependent weight functions to the kernel, and applies this method to posterior probability and KL divergence interpolation estimation.
VaRT: Variational Regression Trees
Sebastian Salazar
Tabular
🎯 What it does: A Variational Regression Tree (VaRT) is proposed—a non-parametric Bayesian model that uses variational inference to approximate the posterior in tree space, applicable for regression and causal inference.
VAST: A Vision-Audio-Subtitle-Text Omni-Modality Foundation Model and Dataset
Sihan Chen (University of Chinese Academy of Sciences), Jing Liu (University of Chinese Academy of Sciences)
GenerationRetrievalTransformerLarge Language ModelContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: A multimodal video subtitle dataset VAST-27M with a scale of 27M has been constructed, and a foundational model VAST capable of perceiving visual, audio, subtitle, and text modalities has been trained.
VCC: Scaling Transformers to 128K Tokens or More by Prioritizing Important Tokens
Zhanpeng Zeng (University of Wisconsin Madison), Shuai Zheng (AWS AI)
CompressionComputational EfficiencyTransformerText
🎯 What it does: This paper proposes the VIP-Token Focused Compression (VCC) scheme, which compresses and decompresses long sequences between Transformer layers, significantly reducing the computational and memory requirements for sequence length.
VeriX: Towards Verified Explainability of Deep Neural Networks
Min Wu (Stanford University), Clark Barrett (Stanford University)
Autonomous DrivingExplainability and InterpretabilityImage
🎯 What it does: This paper proposes VERIX, which generates optimal robust explanations and counterfactuals on decision boundaries based on constraint solving and feature sensitivity ranking.
Versatile Energy-Based Probabilistic Models for High Energy Physics
Taoli Cheng (Mila University of Montreal), Aaron Courville (Mila University of Montreal)
GenerationAnomaly DetectionTransformerContrastive LearningTabularPhysics Related
🎯 What it does: A multi-task framework based on an energy model has been constructed for the generation, anomaly detection, and classification of high-energy physics events.
ViCA-NeRF: View-Consistency-Aware 3D Editing of Neural Radiance Fields
Jiahua Dong (University of Illinois Urbana-Champaign), Yu-Xiong Wang (University of Illinois Urbana-Champaign)
Image TranslationGenerationDiffusion modelNeural Radiance FieldImage
🎯 What it does: By combining text prompts with a two-dimensional diffusion model (Instruct-Pix2Pix) and utilizing NeRF depth information to edit key views, a 3D editing framework called ViCA-NeRF is proposed, which propagates edits to the panorama through projection and mixing methods while maintaining viewpoint consistency.
Video Dynamics Prior: An Internal Learning Approach for Robust Video Enhancements
Gaurav Shrivastava (University of Maryland), Abhinav Shrivastava (University of Maryland)
RestorationSuper ResolutionConvolutional Neural NetworkRecurrent Neural NetworkVideo
🎯 What it does: A novel internal learning framework (Video Dynamics Prior, VDP) is proposed, which utilizes only the statistical information of the video itself without external training data to achieve low-level video processing tasks (denoising, frame interpolation, super-resolution, and object removal).
Video Prediction Models as Rewards for Reinforcement Learning
Alejandro Escontrela (University of California), Pieter Abbeel (University of California)
Robotic IntelligenceTransformerReinforcement LearningGenerative Adversarial NetworkVideo
🎯 What it does: The VIPER algorithm is proposed, which utilizes the likelihood of a pre-trained video prediction model as a reward to train an RL agent to learn complex behaviors from unlabeled videos.
Video-Mined Task Graphs for Keystep Recognition in Instructional Videos
Kumar Ashutosh (University of Texas at Austin), Kristen Grauman (University of Texas at Austin)
RecognitionRepresentation LearningGraph Neural NetworkContrastive LearningVideoMultimodality
🎯 What it does: This paper proposes to automatically mine probabilistic task graphs from unlabeled instructional videos and use these graphs as priors to improve the recognition of key steps and video representation learning.
VideoComposer: Compositional Video Synthesis with Motion Controllability
Xiang Wang (Alibaba Group), Jingren Zhou (Alibaba Group)
GenerationData SynthesisConvolutional Neural NetworkTransformerDiffusion modelVideoText
🎯 What it does: Constructed VideoComposer, a composable video generation framework that supports unified guidance of various conditions including text, spatial (single frame, sketch, style), and temporal (motion vectors, depth, masks, sketch sequences), achieving flexible and controllable video synthesis.
VillanDiffusion: A Unified Backdoor Attack Framework for Diffusion Models
Sheng-Yen Chou (Chinese University of Hong Kong), Tsung-Yi Ho (Chinese University of Hong Kong)
GenerationAdversarial AttackDiffusion modelScore-based ModelImageTextStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: We propose VillanDiffusion, a unified backdoor attack framework capable of injecting image or text-triggered backdoors across various diffusion models (DDPM, LDM, NCSN, etc.) and multiple untrained samplers (DPM-Solver, UniPC, DEIS, etc.).
VisionLLM: Large Language Model is also an Open-Ended Decoder for Vision-Centric Tasks
Wenhai Wang (Chinese University of Hong Kong), Jifeng Dai (Tsinghua University)
Object DetectionSegmentationTransformerLarge Language ModelVision Language ModelImage
🎯 What it does: Proposes the VisionLLM framework, using LLM as an open decoder to accomplish vision-centric tasks.
ViSt3D: Video Stylization with 3D CNN
Ayush Pande (Indian Institute of Technology Kanpur), Gaurav Sharma (TensorTour and Indian Institute of Technology Kanpur)
Image TranslationGenerationConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: The first method for achieving video style transfer using 3D CNN is proposed, generating temporally consistent artistic videos by decoupling appearance and motion information.
Visual Explanations of Image-Text Representations via Multi-Modal Information Bottleneck Attribution
Ying Wang (New York University), Andrew Gordon Wilson (New York University)
Explainability and InterpretabilityTransformerContrastive LearningImageTextMultimodality
🎯 What it does: Proposes an explainability method based on multimodal information bottleneck (M2IB) for generating attribution maps for image-text pairs;
Visual Instruction Inversion: Image Editing via Image Prompting
Thao Nguyen (University of Wisconsin Madison), Yong Jae Lee (University of Wisconsin Madison)
Image TranslationGenerationPrompt EngineeringDiffusion modelImage
🎯 What it does: A method for image editing is proposed through Visual Instruction Inversion, which only requires a pair of 'before and after' example images to learn the editing direction and apply that direction to new images combined with text prompts.
Visual Instruction Tuning
Haotian Liu (University of Wisconsin Madison), Yong Jae Lee (Columbia University)
RecognitionGenerationRetrievalTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: This study proposes LLaVA (Large Language and Vision Assistant), a multimodal model capable of visual question answering and chatting, by connecting the CLIP visual encoder with the Vicuna LLM through linear projection and performing end-to-end instruction tuning on visual instruction-following data generated by GPT-4.
Visual Programming for Step-by-Step Text-to-Image Generation and Evaluation
Jaemin Cho (University of North Carolina at Chapel Hill), Mohit Bansal (University of North Carolina at Chapel Hill)
Object DetectionGenerationTransformerLarge Language ModelSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: Two visual programming frameworks are proposed: VPGEN for interpretable step-by-step text-to-image generation; VPEVAL for interpretable text-to-image evaluation, both implemented through readable programs.
VLATTACK: Multimodal Adversarial Attacks on Vision-Language Tasks via Pre-trained Models
Ziyi Yin (Pennsylvania State University), Fenglong Ma (Pennsylvania State University)
RecognitionAdversarial AttackTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper studies adversarial attacks on downstream fine-tuning models using only publicly available pre-trained vision-language models under black-box conditions, and proposes the VLATTACK method.
Vocabulary-free Image Classification
Alessandro Conti (University of Trento), Elisa Ricci (Fondazione Bruno Kessler)
ClassificationRetrievalTransformerVision Language ModelImageMultimodality
🎯 What it does: Proposes the Vocabulary-free Image Classification (VIC) task and develops the CaSED method to achieve vocabulary-free image classification.
VOCE: Variational Optimization with Conservative Estimation for Offline Safe Reinforcement Learning
Jiayi Guan (Tongji University), changjun jiang
OptimizationSafty and PrivacyReinforcement Learning
🎯 What it does: This paper proposes an algorithm VOCE for offline safe reinforcement learning, which combines a probabilistic inference framework and conservative Q-value estimation to optimize both rewards and satisfy safety constraints on offline datasets.
Voicebox: Text-Guided Multilingual Universal Speech Generation at Scale
Matthew Le (Meta), Wei-Ning Hsu (Meta)
GenerationData SynthesisTransformerFlow-based ModelAudio
🎯 What it does: Voicebox is a large-scale multilingual text-guided non-autoregressive flow model trained on massive audiobook data, achieving multi-task capabilities such as speech filling, zero-shot TTS, denoising, editing, and diverse sampling.
Volume Feature Rendering for Fast Neural Radiance Field Reconstruction
Kang Han (La Trobe University), Lu Yu (La Trobe University)
GenerationComputational EfficiencyNeural Radiance FieldImage
🎯 What it does: This paper studies a volume feature rendering method that allows rendering pixels with only one evaluation of the color network in NeRF reconstruction, thereby accelerating training and inference.
VoxDet: Voxel Learning for Novel Instance Detection
Bowen Li (Carnegie Mellon University), Sebastian Scherer (Carnegie Mellon University)
Object DetectionPoint Cloud
🎯 What it does: This paper proposes a voxel learning-based 3D geometric perception framework called VoxDet, designed for detecting unseen instances under multi-view template conditions.
VPGTrans: Transfer Visual Prompt Generator across LLMs
Ao Zhang (National University of Singapore), Tat-Seng Chua (National University of Singapore)
TransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: A two-stage visual prompt generator (VPG) transfer framework called VPGTrans is proposed, which can efficiently transfer VPG between different LLM sizes and types, significantly reducing training costs.
VPP: Efficient Conditional 3D Generation via Voxel-Point Progressive Representation
Zekun Qi (Xi'an Jiaotong University), Kaisheng Ma (Tsinghua University)
GenerationData SynthesisTransformerGenerative Adversarial NetworkPoint Cloud
🎯 What it does: A 3D generation method based on voxel-point progressive generation (VPP) is proposed, which can efficiently generate multi-category, high-resolution point clouds and supports downstream tasks such as editing and completion.
VRA: Variational Rectified Activation for Out-of-distribution Detection
Mingyu Xu (Chinese Academy of Sciences), Jianhua Tao (Tsinghua University)
Anomaly DetectionConvolutional Neural NetworkTransformerImage
🎯 What it does: A post-hoc OOD detection method called VRA (Variational Rectified Activation) is proposed, which designs an activation function through variational methods to suppress abnormally low/high activations and amplify intermediate activations.
Vulnerabilities in Video Quality Assessment Models: The Challenge of Adversarial Attacks
Aoxiang Zhang, Yuan-Gen Wang (Guangzhou University)
Adversarial AttackVideo
🎯 What it does: A robustness evaluation of no-reference video quality assessment (NR-VQA) models is conducted, systematically studying their vulnerability to adversarial attacks for the first time, and proposing a white-box attack method based on Score-Reversed Boundary Loss and a black-box attack method based on patch random search.
WalkLM: A Uniform Language Model Fine-tuning Framework for Attributed Graph Embedding
Yanchao Tan (Fuzhou University), Carl Yang (Emory University)
ClassificationRepresentation LearningGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningGraphBiomedical DataElectronic Health Records
🎯 What it does: The WalkLM framework is proposed, which generates attribute-rich random walk text sequences on attribute graphs to perform unsupervised fine-tuning of pre-trained language models, resulting in unified node embeddings.
Wasserstein distributional robustness of neural networks
Xingjian Bai (University of Oxford), Jan Obloj
OptimizationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a framework based on Wasserstein Distributionally Robust Optimization (W-DRO) to unify the study of adversarial attacks and robust training of neural networks, and extends it to distributed threat models.
Wasserstein Gradient Flows for Optimizing Gaussian Mixture Policies
Hanna Ziesche (Bosch Center for Artificial Intelligence), Leonel Rozo (Bosch Center for Artificial Intelligence)
OptimizationRobotic IntelligenceReinforcement LearningMixture of ExpertsSequential
🎯 What it does: A Gaussian Mixture Model (GMM) strategy optimization method based on Wasserstein gradient flow is proposed, utilizing the Bures-Wasserstein manifold for Riemannian optimization to achieve adaptive updates of existing GMM strategies in robotic motion tasks.
Wasserstein Quantum Monte Carlo: A Novel Approach for Solving the Quantum Many-Body Schrödinger Equation
Kirill Neklyudov (Vector Institute), Alireza Makhzani (Vector Institute)
OptimizationComputational EfficiencyDrug DiscoveryTransformerTabularSequentialPhysics Related
🎯 What it does: A quantum Monte Carlo algorithm based on the Wasserstein metric (WQMC) is proposed and implemented. It treats energy minimization as a gradient flow in the space of Born distributions and projects this flow onto a parameterized deep network to solve the ground state of the many-body Schrödinger equation.
Waypoint Transformer: Reinforcement Learning via Supervised Learning with Intermediate Targets
Anirudhan Badrinath (Stanford University), Emma Brunskill (Stanford University)
TransformerSupervised Fine-TuningReinforcement LearningMultimodality
🎯 What it does: This paper proposes the Waypoint Transformer (WT), which guides policy learning by using intermediate goals (waypoints) within an offline reinforcement learning supervised learning framework.
Weakly Coupled Deep Q-Networks
Ibrahim El Shar (Hitachi America Ltd.), Daniel R. Jiang (Meta)
Reinforcement Learning
🎯 What it does: This paper proposes a reinforcement learning algorithm utilizing weakly coupled structures—Weakly Coupled Deep Q-Network (WCDQN) and its tabular version Weakly Coupled Q-Learning (WCQL). It splits the complete MDP into multiple subproblems and uses Lagrangian relaxation to obtain an upper bound, guiding the convergence of the main Q-learning.
Weakly Supervised 3D Open-vocabulary Segmentation
Kunhao Liu (Nanyang Technological University), Shijian Lu (Nanyang Technological University)
SegmentationKnowledge DistillationNeural Radiance FieldContrastive LearningImage
🎯 What it does: Through weak supervision, knowledge distillation from CLIP and DINO to NeRF is achieved for open vocabulary segmentation of 3D scenes.
Weakly-Supervised Audio-Visual Segmentation
Shentong Mo (Carnegie Mellon University), Bhiksha Raj (Carnegie Mellon University)
SegmentationConvolutional Neural NetworkContrastive LearningVideoMultimodalityAudio
🎯 What it does: A weakly supervised audio-visual segmentation framework WS-AVS is proposed, which trains an audio-visual segmentation model without relying on pixel-level masks using instance-level annotations.
Weakly-Supervised Concealed Object Segmentation with SAM-based Pseudo Labeling and Multi-scale Feature Grouping
Chunming He (Shenzhen International Graduate School Tsinghua University), Xiu Li (Shenzhen International Graduate School Tsinghua University)
Object DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: Utilize SAM to generate pseudo-masks under sparse annotations (points/strokes) and train a weakly supervised hidden object segmentation model through multi-scale feature grouping (MFG).
Weighted ROC Curve in Cost Space: Extending AUC to Cost-Sensitive Learning
Huiyang Shao (Chinese Academy of Sciences), Qingming Huang (Chinese Academy of Sciences)
ClassificationRecommendation SystemAnomaly DetectionOptimizationData-Centric LearningReinforcement LearningTabularFinance Related
🎯 What it does: A long-tail cost-sensitive learning framework is proposed that treats costs as sampleable data, constructing a bi-level optimization model for weighted AUC (WAUC) and cost functions. A differentiable WAUC estimator and convex internal cost optimization are provided, ultimately leading to the SACCL stochastic gradient algorithm for end-to-end training.
Weitzman's Rule for Pandora's Box with Correlations
Evangelia Gergatsouli (University of Wisconsin Madison), Christos Tzamos (University of Wisconsin Madison)
🎯 What it does: A general variant of the Weitzman rule for the Pandora's Box problem with correlated distributions is proposed, along with an improved approximation algorithm and its properties that can be learned through samples.
What can a Single Attention Layer Learn? A Study Through the Random Features Lens
Hengyu Fu (Peking University), Song Mei (University of California Berkeley)
TransformerSequential
🎯 What it does: This paper studies the expressiveness and generalization performance of a single-layer attention layer from the perspective of random features, proving that it can approximate a wide range of permutation-invariant objective functions for key vectors and providing an upper bound on sample complexity.
What Can We Learn from Unlearnable Datasets?
Pedro Sandoval-Segura (University of Maryland), Tom Goldstein (University of Maryland)
Adversarial AttackData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: This study evaluates the vulnerable 'non-learnable' datasets, revealing their impact on deep learning models and proposing new cracking methods.
What Distributions are Robust to Indiscriminate Poisoning Attacks for Linear Learners?
Fnu Suya (University of Virginia), David Evans (University of Virginia)
OptimizationAdversarial AttackConvolutional Neural NetworkImageTabular
🎯 What it does: The study investigates the robustness of linear learners against indiscriminate data poisoning attacks under no defense conditions, revealing the relationship between data distribution and poisoning effects from both theoretical and empirical perspectives.
What Do Deep Saliency Models Learn about Visual Attention?
Shi Chen (University of Minnesota), Qi Zhao (University of Minnesota)
Explainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: An interpretable framework has been developed to decompose the implicit features of deep saliency models into trainable bases, and to quantitatively measure the positive and negative contributions of each semantic to saliency prediction through probability mapping and semantic alignment, thereby enabling a systematic analysis of model behavior.
What functions can Graph Neural Networks compute on random graphs? The role of Positional Encoding
Nicolas Keriven (Centre National de la Recherche Scientifique), Samuel Vaiter (Centre National de la Recherche Scientifique)
Graph Neural NetworkGraph
🎯 What it does: This study investigates the expressible function space of graph neural networks (GNNs) on large random graphs for node tasks and analyzes the impact of positional encoding on expressiveness.
What is Flagged in Uncertainty Quantification? Latent Density Models for Uncertainty Categorization
Hao Sun (University of Cambridge), Mihaela van der Schaar (University of Cambridge)
ClassificationAnomaly DetectionImageTabular
🎯 What it does: Proposes the DAUC framework, which uses a confusion density matrix to classify suspected samples marked by UQ methods into three categories (OOD, Bnd, IDM).
What is the Inductive Bias of Flatness Regularization? A Study of Deep Matrix Factorization Models
Khashayar Gatmiry (Massachusetts Institute of Technology), Ching-Yao Chuang (Massachusetts Institute of Technology)
OptimizationTabular
🎯 What it does: This paper studies the inductive bias of minimizing the Hessian trace in deep linear networks, particularly in the context of learning from linear measurements.
What Knowledge Gets Distilled in Knowledge Distillation?
Utkarsh Ojha (University of Wisconsin Madison), Yong Jae Lee (University of Wisconsin Madison)
Domain AdaptationKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: Investigate which implicit attributes of the teacher network are inherited by the student network during the knowledge distillation process, such as localization, adversarial robustness, data invariance, and consensus on unseen domains.
What Makes Data Suitable for a Locally Connected Neural Network? A Necessary and Sufficient Condition Based on Quantum Entanglement.
Yotam Alexander (Tel Aviv University), Nadav Cohen (Tel Aviv University)
Convolutional Neural NetworkRecurrent Neural NetworkImageTabularPhysics RelatedAudio
🎯 What it does: This paper presents a necessary and sufficient theoretical condition: if a data distribution has low quantum entanglement under certain classical partitions of its features, then the distribution is suitable for local connection neural networks (such as convolutional networks, S4 recurrent networks, and local self-attention networks); otherwise, it is not suitable.
What Makes Good Examples for Visual In-Context Learning?
Yuanhan Zhang (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
Object DetectionSegmentationContrastive LearningImage
🎯 What it does: This study investigates how to enhance context learning effects in visual large models by automatically retrieving good examples, proposing both unsupervised and supervised retrieval frameworks.
What Planning Problems Can A Relational Neural Network Solve?
Jiayuan Mao (Massachusetts Institute of Technology), Leslie Pack Kaelbling (Massachusetts Institute of Technology)
OptimizationGraph Neural NetworkTransformerReinforcement LearningGraph
🎯 What it does: This study investigates the circuit complexity of target conditional strategies in classical discrete planning problems, proposing an analysis framework based on regression width and providing a constructive method for compiling relational neural networks (RelNN).
What Truly Matters in Trajectory Prediction for Autonomous Driving?
Tran Phong, David Hsu (National University of Singapore)
Autonomous DrivingComputational EfficiencyRecurrent Neural NetworkTime Series
🎯 What it does: This paper systematically studies the dynamic gap between static evaluation and real driving through interactive evaluation of predictors in a simulation environment, and proposes a task-driven evaluation protocol.
What You See is What You Read? Improving Text-Image Alignment Evaluation
Michal Yarom (Google Research), Idan Szpektor (Google Research)
Data SynthesisRetrievalTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: This study investigates how to automatically determine whether images and texts are semantically aligned, and proposes a multi-source, multi-task image-text alignment evaluation benchmark named SeeTRUE, along with two automatic evaluation methods (VQ2 and VNLI) to measure the quality of image-text matching.
What’s Left? Concept Grounding with Logic-Enhanced Foundation Models
Joy Hsu (Stanford University), Jiajun Wu (Stanford University)
Robotic IntelligenceTransformerLarge Language ModelImagePoint CloudTime Series
🎯 What it does: A unified Logic-Enhanced Foundation Model (LEFT) is proposed, which generates first-order logic programs through LLM and combines them with domain-specific concept representations in a differentiable executor, achieving concept learning and reasoning across 2D, 3D, time series, and robotic operations.
When are ensembles really effective?
Ryan Theisen (University of California), Michael W. Mahoney (University of California)
ClassificationTransformerTabular
🎯 What it does: The effectiveness of ensemble methods in classification tasks has been studied and verified, and the theoretical relationship between Ensemble Improvement Rate (EIR) and Discrepancy-Error Ratio (DER) has been proposed.
When can Regression-Adjusted Control Variate Help? Rare Events, Sobolev Embedding and Minimax Optimality
Jose Blanchet (Stanford University), Lexing Ying (Stanford University)
Optimization
🎯 What it does: This paper studies the application of machine learning-based estimators as control variables to reduce the variance of Monte Carlo sampling and explores the key factors affecting the efficiency of control variables.
When Can We Track Significant Preference Shifts in Dueling Bandits?
Joe Suk (Columbia University), Arpit Agarwal (Columbia University)
Recommendation SystemOptimization
🎯 What it does: The study investigates preference changes in the K-armed adversarial bandit problem and proposes an adaptive algorithm to achieve dynamic regret of O(√KLT˜), where ˜L is the number of significant preference changes.
When Demonstrations meet Generative World Models: A Maximum Likelihood Framework for Offline Inverse Reinforcement Learning
Siliang Zeng (University of Minnesota), Mingyi Hong (University of Minnesota)
Robotic IntelligenceReinforcement LearningWorld ModelSequential
🎯 What it does: This paper proposes an offline inverse reinforcement learning framework based on maximum likelihood estimation, which jointly learns the reward function and optimal policy using a world model and a conservative policy.
When Do Graph Neural Networks Help with Node Classification? Investigating the Homophily Principle on Node Distinguishability
Sitao Luan (McGill University), Doina Precup (McGill University)
ClassificationGraph Neural NetworkGraph
🎯 What it does: This paper systematically studies the impact of the homophily principle on the performance of Graph Neural Networks (GNN) in node classification tasks, proposing a new graph generation model CSBM-H and designing two metrics to measure node distinguishability (ND) — Probabilistic Bayes Error (PBE) and Negative Generalized Jeffreys Divergence (DNGJ). Through theoretical derivation and experimental validation, it proves that the advantages of GNN are closely related to the statistical feature that 'the embedding distance of similar nodes is less than that of dissimilar nodes', and based on this, it proposes a classifier-based performance metric (CPM) that can directly provide a significance threshold without training.
When Do Transformers Shine in RL? Decoupling Memory from Credit Assignment
Tianwei Ni (Mila Université de Montréal), Pierre-Luc Bacon (Mila Université de Montréal)
Recurrent Neural NetworkTransformerReinforcement LearningSequential
🎯 What it does: This paper systematically analyzes the memory and credit assignment capabilities of Transformers in reinforcement learning, proposes formal definitions for memory length and credit assignment length, and designs configurable toy tasks (such as Passive/Active T-Maze) to decouple these two capabilities. It then evaluates Transformer-based and LSTM-based RL algorithms in various POMDP environments (including custom T-Maze, Passive/Active Visual Match, Key-to-Door, PyBullet Benchmarks, etc.) to explore their performance under different memory/credit assignment lengths.
When Does Confidence-Based Cascade Deferral Suffice?
Wittawat Jitkrittum (Google Research), Sanjiv Kumar (Google Research)
ClassificationDomain AdaptationImage
🎯 What it does: This paper studies when confidence-based deferral rules can achieve optimality in model cascades, proposes a Bayesian optimal deferral rule, and discusses its shortcomings in special cases; it then introduces a post-hoc deferral learning method that does not require invoking subsequent models during inference, and validates its superiority in three practical scenarios (expert model specialization, label noise, and distribution shift).
When Does Optimizing a Proper Loss Yield Calibration?
Jarosław Błasiok (Columbia University), Preetum Nakkiran (Apple)
Optimization
🎯 What it does: This study investigates when good calibration results can be achieved when optimizing strictly proper loss functions within a finite family of functions, and provides a formal theoretical definition.
When is Agnostic Reinforcement Learning Statistically Tractable?
Zeyu Jia (Massachusetts Institute of Technology), Nathan Srebro (Toyota Technological Institute at Chicago)
Reinforcement Learning
🎯 What it does: This paper studies the sample complexity of agnostic reinforcement learning (RL) under large state spaces without model assumptions, proposing a spanning capacity based on policy classes to measure learning difficulty, and proving that it is both necessary and sufficient under generative models;
When Visual Prompt Tuning Meets Source-Free Domain Adaptive Semantic Segmentation
Xinhong Ma (Huawei Noah's Ark Lab), Yunhe Wang (Huawei Noah's Ark Lab)
SegmentationDomain AdaptationTransformerPrompt EngineeringImage
🎯 What it does: A visual prompt tuning framework for source-free domain adaptive semantic segmentation, Uni-UVPT, is proposed, which adapts a frozen Transformer model using only a small number of learnable parameters.
Where are we in the search for an Artificial Visual Cortex for Embodied Intelligence?
Arjun Majumdar (Georgia Tech), Franziska Meier (Meta AI)
Robotic IntelligenceTransformerReinforcement LearningImageVideoBenchmark
🎯 What it does: A comprehensive systematic evaluation of visual pre-trained representations (PVR) was conducted across 17 different embodied AI tasks, leading to the construction of the CORTEXBENCH benchmark. Additionally, the MAE pre-trained Vision Transformer was used on over 5.6 million frames of egocentric video and the ImageNet dataset to obtain the VC-1 model, which was then task-specifically adapted and validated on real hardware.
Where Did I Come From? Origin Attribution of AI-Generated Images
Zhenting Wang (Rutgers University), Shiqing Ma (Sony AI)
GenerationData SynthesisOptimizationImage
🎯 What it does: A model-agnostic AI image source attribution method has been developed without modifications, utilizing reverse engineering to reconstruct inputs and distinguishing whether an image was generated by a specific model through reconstruction loss;
Where2Explore: Few-shot Affordance Learning for Unseen Novel Categories of Articulated Objects
Chuanruo Ning (Peking University), Hao Dong (Peking University)
Robotic IntelligencePoint Cloud
🎯 What it does: The Where2Explore framework is proposed, utilizing cross-category local geometric similarity to achieve a small amount of interactive exploration and learn operability for unseen categories.
Which Models have Perceptually-Aligned Gradients? An Explanation via Off-Manifold Robustness
Suraj Srinivas (Harvard University), Himabindu Lakkaraju (Harvard University)
ClassificationRecognitionDiffusion modelImage
🎯 What it does: This study investigates the phenomenon of perceptual alignment of robust model gradients, proposing discrete manifold robustness to explain why gradients lie on the signal manifold, and introduces signal-interference decomposition.
White-Box Transformers via Sparse Rate Reduction
Yaodong Yu (University of California), Yi Ma (University of California)
ClassificationOptimizationExplainability and InterpretabilityTransformerImage
🎯 What it does: A fully white-box Transformer architecture called CRATE is proposed, which interprets self-attention as denoising/compression of low-dimensional subspace and MLP as ISTA sparse coding, forming a set of interpretable and trainable hierarchical networks;
Why Did This Model Forecast This Future? Information-Theoretic Saliency for Counterfactual Explanations of Probabilistic Regression Models
Chirag Raman (Delft University of Technology), Marco Loog (Delft University of Technology)
Explainability and InterpretabilityTime SeriesSequential
🎯 What it does: A post-hoc explanation framework is proposed, utilizing pre-attentive saliency from information theory to identify significant moments in the predictions of probabilistic multivariate time series regression models, supporting causal explanations.
Why Does Sharpness-Aware Minimization Generalize Better Than SGD?
Zixiang Chen (University of California), Quanquan Gu (University of California)
OptimizationConvolutional Neural NetworkTabular
🎯 What it does: This paper compares two training methods—Standard Stochastic Gradient Descent (SGD) and Sharpness-Aware Minimization (SAM)—in terms of generalization performance on a two-layer convolutional ReLU network through theoretical analysis and experimental validation, revealing the advantages of SAM in preventing noise learning and achieving benign overfitting.
Why think step by step? Reasoning emerges from the locality of experience
Ben Prystawski (Stanford University), Noah Goodman
TransformerLarge Language ModelChain-of-Thought
🎯 What it does: The effectiveness of chain reasoning in language models was studied, and it was demonstrated that the local structure of training data allows intermediate reasoning steps to reduce bias.
Wide Neural Networks as Gaussian Processes: Lessons from Deep Equilibrium Models
Tianxiang Gao (Iowa State University), Hongyang Gao (Iowa State University)
Gaussian SplattingImage
🎯 What it does: This paper analyzes the Deep Equilibrium Model (DEQ), proving that as the width approaches infinity, its output converges to a Gaussian process, and the corresponding kernel function is strictly positive definite;
Window-Based Distribution Shift Detection for Deep Neural Networks
Guy Bar-Shalom (Technion Israel Institute of Technology), Ran El-Yaniv (Technion Israel Institute of Technology)
ClassificationDomain AdaptationAnomaly DetectionConvolutional Neural NetworkTransformerImage
🎯 What it does: A coverage-based window distribution shift detection method (Coverage-Based Detection, CBD) is proposed, which monitors whether the input stream undergoes distribution changes by calculating the lower bound of coverage based on the confidence (such as entropy) of the pre-trained model.