NeurIPS 2023 Papers — Page 17
Conference on Neural Information Processing Systems · 3218 papers
Latent Space Translation via Semantic Alignment
Valentino Maiorca (Sapienza University of Rome), Emanuele Rodolà (Sapienza University of Rome)
Domain AdaptationRepresentation LearningConvolutional Neural NetworkTransformerImageTextMultimodality
🎯 What it does: This paper proposes a method to directly estimate and apply a transformation T to translate between the latent spaces of pre-trained models, achieving zero-shot model stitching across architectures, modalities, and tasks without the need to retrain the decoder.
Laughing Hyena Distillery: Extracting Compact Recurrences From Convolutions
Stefano Massaroli (Mila and University of Montreal), Yoshua Bengio (Mila and University of Montreal)
CompressionComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkLarge Language ModelText
🎯 What it does: After pre-training, compress the long convolutional sequence model to extract a low-dimensional state space model for recursive inference;
Layer-Neighbor Sampling --- Defusing Neighborhood Explosion in GNNs
Muhammed Fatih Balin, Umit Catalyurek
Graph Neural NetworkGraph
🎯 What it does: This paper proposes a new sampling algorithm called LABOR (Layer-Neighbor Sampling), which combines the advantages of neighbor sampling and hierarchical sampling, and achieves unbiased estimation through Poisson sampling.
LayoutGPT: Compositional Visual Planning and Generation with Large Language Models
Weixi Feng (University of California), William Yang Wang (University of California)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringImageMultimodality
🎯 What it does: Utilizing large language models (LLM) to generate visual layouts of 2D images and 3D indoor scenes through in-context learning and CSS-style prompts, achieving high-quality visual content generation in collaboration with existing layout-to-image generation models.
LayoutPrompter: Awaken the Design Ability of Large Language Models
Jiawei Lin (Xi'an Jiaotong University), Dongmei Zhang (Microsoft)
GenerationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A training-agnostic method for conditional layout generation using large language models, called LayoutPrompter, is proposed.
LD2: Scalable Heterophilous Graph Neural Network with Decoupled Embeddings
Ningyi Liao (Nanyang Technological University), Jieming Shi (Hong Kong Polytechnic University)
Graph Neural NetworkGraph
🎯 What it does: This paper proposes a scalable heterogeneous graph neural network LD2, which utilizes precomputed low-dimensional adjacency and feature embeddings, completely decoupling graph structure propagation during training, and supports lightweight mini-batch training for large-scale heterogeneous graphs.
LEACE: Perfect linear concept erasure in closed form
Nora Belrose (EleutherAI), Stella Biderman (Booz Allen Hamilton)
TransformerLarge Language ModelText
🎯 What it does: The LEACE (Least-Squares Concept Erasure) method is proposed, which can completely eliminate specified concepts (such as gender or part of speech) from internal representations through linear transformations without changing model parameters.
Learn to Categorize or Categorize to Learn? Self-Coding for Generalized Category Discovery
Sarah Rastegar (University of Amsterdam), Cees G. M. Snoek (University of Amsterdam)
ClassificationOptimizationRepresentation LearningAuto EncoderContrastive LearningImage
🎯 What it does: This study investigates how to discover unknown categories during testing using the InfoSieve method.
Learning a 1-layer conditional generative model in total variation
Ajil Jalal (University of California Berkeley), Eric Price (University of Texas Austin)
GenerationData SynthesisTabular
🎯 What it does: This paper studies how to learn conditional generative models with very few samples without relying on input distribution assumptions, focusing on proving that a single-layer ReLU generative model can achieve a given accuracy using maximum likelihood estimation (MLE) in terms of total variation distance.
Learning a Neuron by a Shallow ReLU Network: Dynamics and Implicit Bias for Correlated Inputs
Dmitry Chistikov (University of Warwick), Ranko Lazic (University of Warwick)
Tabular
🎯 What it does: This paper analyzes the training dynamics of a single hidden layer ReLU network learning a single neuron, starting from small initialization and using gradient flow, proving convergence to zero loss and an implicit bias towards minimal rank.
Learning Adaptive Tensorial Density Fields for Clean Cryo-ET Reconstruction
YUANHAO WANG, Wolfgang Heidrich (King Abdullah University of Science and Technology)
RestorationData SynthesisImage
🎯 What it does: The paper proposes a deep learning framework based on adaptive quadtree and tensor decomposition for the rapid reconstruction and denoising of 3D density fields from tilted series Cryo-ET data.
Learning Adversarial Low-rank Markov Decision Processes with Unknown Transition and Full-information Feedback
Canzhe Zhao (Shanghai Jiao Tong University), Shuai Li (Shanghai Jiao Tong University)
OptimizationReinforcement Learning
🎯 What it does: A strategy optimization algorithm named POLO is proposed to achieve a sublinear regret upper bound in low-rank Markov decision processes with unknown transitions and adversarial losses.
Learning and Collusion in Multi-unit Auctions
Simina Branzei, Yanjun Han (New York University)
🎯 What it does: This paper studies the learning and collusion issues in multi-unit uniform price auctions, proposing an offline optimal bidding scheme and an online low-regret learning algorithm, and analyzes the equilibrium and collusion characteristics of two price formats.
Learning and processing the ordinal information of temporal sequences in recurrent neural circuits
Xiaolong Zou (Qiyuan Lab), Yuanyuan Mi (Tsinghua University)
RecognitionData SynthesisRecurrent Neural NetworkTime SeriesSequentialAudio
🎯 What it does: This study investigates how recurrent neural networks (RNNs) learn and represent ordinal information in time series, utilizing tree-structured absorbing dynamics to achieve separation and binding of sequence and content.
Learning better with Dale’s Law: A Spectral Perspective
Pingsheng Li (McGill University), Blake Aaron Richards
Recurrent Neural NetworkSequential
🎯 What it does: The study trains RNNs (ColEI and DANN) while maintaining Dale's principle and analyzes their differences in performance.
Learning Better with Less: Effective Augmentation for Sample-Efficient Visual Reinforcement Learning
Guozheng Ma (Tsinghua University), Dacheng Tao (University of Sydney)
Autonomous DrivingReinforcement LearningImage
🎯 What it does: This paper systematically analyzes the key attributes of data augmentation (DA) in visual reinforcement learning through experiments, and based on this, proposes a new random PadResize (Rand PR) transformation and a periodic fusion augmentation scheme (CycAug) to enhance sample efficiency.
Learning Causal Models under Independent Changes
Sarah Mameche (CISPA Helmholtz Center for Information Security), Jilles Vreeken (CISPA Helmholtz Center for Information Security)
TabularTime Series
🎯 What it does: In multi-environment non-i.i.d. data, a computable MDL score is constructed using the principle of algorithmic independence and Gaussian process (GP) models, thereby learning causal mechanisms and determining causal directions;
Learning Curves for Deep Structured Gaussian Feature Models
Jacob A Zavatone-Veth, Cengiz Pehlevan (Harvard University)
🎯 What it does: This paper studies the learning curve of deep linear random feature models (Deep Linear RFMs) and explores the impact of weight correlation (weight heterogeneity) on generalization error.
Learning Curves for Noisy Heterogeneous Feature-Subsampled Ridge Ensembles
Benjamin Samuel Ruben, Cengiz Pehlevan (Harvard University)
Image
🎯 What it does: A feature sub-sampling ridge regression ensemble method under noise is proposed and analyzed, providing an analyzable learning curve and simplifying under the assumption of equal correlated features.
Learning Cuts via Enumeration Oracles
Daniel Thuerck (Quantagonia), Sebastian Pokutta (Zuse Institute Berlin)
OptimizationTabular
🎯 What it does: A local cutting learning framework without LP is proposed, which utilizes the Frank-Wolfe algorithm and an enumeration oracle to generate cuts in a low-dimensional subspace, and then maps them back to the original space through lifting; its effectiveness is validated on the multidimensional knapsack problem (MKP).
Learning DAGs from Data with Few Root Causes
Panagiotis Misiakos (ETH Zurich), Markus Püschel (ETH Zurich)
OptimizationGraph
🎯 What it does: A continuous optimization method named SparseRC is proposed for learning the sparse causal linear structural equation model (SEM) that generates directed acyclic graphs (DAGs);
Learning Dense Flow Field for Highly-accurate Cross-view Camera Localization
Zhenbo Song (Nanjing University of Science and Technology), Yujiao Shi (ShanghaiTech University)
Pose EstimationAutonomous DrivingConvolutional Neural NetworkOptical FlowImage
🎯 What it does: An end-to-end dense pixel-level optical flow learning framework is proposed, utilizing the matching of ground image projections to bird's-eye view (BEV) and satellite images to achieve precise localization of the ground camera's 3-DoF pose.
Learning Descriptive Image Captioning via Semipermeable Maximum Likelihood Estimation
Zihao Yue (Renmin University of China), Qin Jin (Renmin University of China)
GenerationRetrievalTransformerSupervised Fine-TuningVision Language ModelImageVideoText
🎯 What it does: A new training objective called SMILE (Semipermeable Maximum Likelihood Estimation) is proposed, which normalizes probabilities only for the current set of label words at each time step, suppressing the 'concise optimization' caused by MLE and encouraging the generation of more detailed image descriptions; further training and evaluation are conducted on the BLIP model.
Learning Dictionary for Visual Attention
Yingjie Liu (East China Normal University), Xian Wei (East China Normal University)
ClassificationSegmentationTransformerImagePoint Cloud
🎯 What it does: A visual attention module based on dictionary learning, Dic-Attn, is proposed, which utilizes sparse coding to decompose features into dictionaries and sparse representations, and dynamically weights the reconstruction of attention maps in both channel and spatial domains.
Learning Domain-Aware Detection Head with Prompt Tuning
Haochen Li (Intelligent Software Research Center, Institute of Software, Chinese Academy of Sciences), Yunji Chen (State Key Lab of Processors, Institute of Computing Technology, Chinese Academy of Sciences)
Object DetectionDomain AdaptationTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes a method that utilizes a Visual-Language Model (VLM) as a robust detection backbone and generates detection heads for each domain through learnable domain adaptation prompts to achieve adaptive learning for cross-domain object detection.
Learning Dynamic Attribute-factored World Models for Efficient Multi-object Reinforcement Learning
Fan Feng (City University of Hong Kong), Sara Magliacane (University of Amsterdam)
Recurrent Neural NetworkReinforcement LearningWorld ModelSequential
🎯 What it does: This paper studies the Dynamic Attribute Factorized World Model (DAFT-RL), achieving cross-environment and cross-object attribute generalization in multi-object reinforcement learning.
Learning Efficient Coding of Natural Images with Maximum Manifold Capacity Representations
Thomas Edward Yerxa (New York University), SueYeon Chung (New York University)
Object DetectionRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A self-supervised learning method based on maximum manifold capacity (MMCR) is proposed, which simplifies the manifold capacity theory into a nuclear norm loss to directly optimize the linear separability and compressibility of representations.
Learning Efficient Surrogate Dynamic Models with Graph Spline Networks
Chuanbo Hua (Korea Advanced Institute of Science and Technology), Jinkyoo Park (Korea Advanced Institute of Science and Technology)
Graph Neural NetworkGraphPhysics Related
🎯 What it does: This paper proposes GRAPHSPLINENETS, which quickly predicts the dynamics of continuous physical systems through graph neural networks and orthogonal spline interpolation (OSC).
Learning Energy-based Model via Dual-MCMC Teaching
Jiali Cui (Stevens Institute of Technology), Tian Han (Stevens Institute of Technology)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: A joint learning framework of dual MCMC teaching is proposed, which jointly trains the energy-based model (EBM), generator model, and inference model to match the empirical distribution and each other's distribution under maximum likelihood conditions, thereby achieving high-quality EBM learning.
Learning Energy-Based Prior Model with Diffusion-Amortized MCMC
Peiyu Yu (University of California Los Angeles), Ying Nian Wu (University of California Los Angeles)
GenerationAnomaly DetectionKnowledge DistillationDiffusion modelAuto EncoderImage
🎯 What it does: A long-term MCMC amortization method based on diffusion models (DAMC) is proposed for learning latent space energy-based models (LEBM), addressing the issue of low sampling quality in traditional short-run MCMC.
Learning Environment-Aware Affordance for 3D Articulated Object Manipulation under Occlusions
Ruihai Wu (Peking University), Hao Dong (Peking University)
Robotic IntelligenceContrastive LearningPoint Cloud
🎯 What it does: This study investigates how to utilize environment-aware affordance to predict point-level operability when robots manipulate three-dimensional joint objects in occluded environments.
Learning Exponential Families from Truncated Samples
Jane Lee, Manolis Zampetakis (Yale University)
OptimizationComputational Efficiency
🎯 What it does: An efficient algorithm is proposed for learning generalized exponential family distributions from truncated samples.
Learning Fine-grained View-Invariant Representations from Unpaired Ego-Exo Videos via Temporal Alignment
Zihui Xue (University of Texas at Austin), Kristen Grauman (University of Texas at Austin)
RetrievalRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningVideoBenchmark
🎯 What it does: Learn fine-grained action representations from unpaired forward and backward videos using a time-aligned self-supervised objective, achieving representation invariance across two perspectives.
Learning from Active Human Involvement through Proxy Value Propagation
Zhenghao Peng (University of California), Bolei Zhou (University of Edinburgh)
Autonomous DrivingReinforcement Learning from Human FeedbackReinforcement LearningSequential
🎯 What it does: A Proxy Value Propagation (PVP) method is proposed, allowing reinforcement learning to align behavior directly through human active intervention and demonstration learning without using a reward function.
Learning From Biased Soft Labels
Hua Yuan (Southeast University), Yong Rui (Lenovo Research)
ClassificationKnowledge DistillationImage
🎯 What it does: This paper studies how a teacher model can effectively train a student model in the presence of significant soft label bias, and proposes two metrics to measure the effectiveness of soft labels.
Learning from Both Structural and Textual Knowledge for Inductive Knowledge Graph Completion
Kunxun Qi (Sun Yat-sen University), Hai Wan (Sun Yat-sen University)
Recurrent Neural NetworkTransformerSupervised Fine-TuningTextGraph
🎯 What it does: This paper proposes a two-stage LSTK framework, which first uses distance supervision and a text entailment model to extract soft triples from text corpora, and then uses these soft triples along with the original structured triples to learn text enhancement rules (TE-rules) for inductive reasoning in knowledge graphs.
Learning from Rich Semantics and Coarse Locations for Long-tailed Object Detection
Lingchen Meng (Fudan University), Yu-Gang Jiang (Fudan University)
Object DetectionTransformerContrastive LearningImage
🎯 What it does: By leveraging the rich semantic information and coarse location (whole image box) from additional image classification data to enhance the training of long-tail object detectors, the detection model achieves better recognition capabilities for low-sample categories.
Learning from Visual Observation via Offline Pretrained State-to-Go Transformer
Bohan Zhou (Peking University), Zongqing Lu (Peking University)
TransformerReinforcement LearningGenerative Adversarial NetworkVideo
🎯 What it does: A two-stage framework is proposed: first, an offline pre-trained State-to-Go Transformer is used to predict the potential transitions of visual observation sequences, and an intrinsic reward is generated through a WGAN discriminator, followed by training the policy using only these rewards in online reinforcement learning.
Learning Functional Transduction
Mathieu Chalvidal (Capital Fund Management), Rufin VanRullen (Centre de Recherche Cerveau et Cognition)
OptimizationMeta LearningTransformerTabularTime SeriesPhysics Related
🎯 What it does: This paper proposes a Transducer model based on the theory of vector-valued reproducing kernel Banach spaces, which instantaneously completes the regression from data to unknown functions in a single forward pass;
Learning Generalizable Agents via Saliency-guided Features Decorrelation
Sili Huang (Jilin University), Bo Yang (Jilin University)
Robotic IntelligenceReinforcement LearningVideo
🎯 What it does: A sample weighting method based on Saliency-Guided Features Decorrelation (SGFD) is proposed to enhance the generalization ability of visual reinforcement learning agents in unseen environments.
Learning in the Presence of Low-dimensional Structure: A Spiked Random Matrix Perspective
Jimmy Ba (University of Toronto), Denny Wu (New York University)
Tabular
🎯 What it does: The study investigates the feasibility of learning single-index models under high-dimensional data with low-dimensional structures;
Learning Interpretable Low-dimensional Representation via Physical Symmetry
Xuanjie Liu (Mohamed bin Zayed University of Artificial Intelligence), Gus Xia (Mohamed bin Zayed University of Artificial Intelligence)
Explainability and InterpretabilityRepresentation LearningAuto EncoderVideoTime SeriesPhysics RelatedAudio
🎯 What it does: The paper proposes using physical symmetry as a prior constraint for self-supervised learning to learn interpretable low-dimensional temporal representations.
Learning Invariant Molecular Representation in Latent Discrete Space
Xiang Zhuang (Zhejiang University), Huajun Chen (Zhejiang Lab)
Representation LearningDrug DiscoveryGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A new framework for learning molecular invariant representations in discrete latent spaces is proposed (iMoLD).
Learning Invariant Representations of Graph Neural Networks via Cluster Generalization
Donglin Xia (Beijing University of Posts and Telecommunications), Chuan Shi (Beijing University of Posts and Telecommunications)
Representation LearningGraph Neural NetworkGraph
🎯 What it does: This paper addresses the issue of GNN generalization decline caused by changes in graph structure and proposes the Cluster Information Transfer (CIT) mechanism to learn node representations that are robust to structural drift.
Learning Invariant Representations with a Nonparametric Nadaraya-Watson Head
Alan Q. Wang (Cornell University), Mert R. Sabuncu (Cornell University)
ClassificationDomain AdaptationRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a non-parametric domain-invariant representation learning based on the Nadaraya-Watson (NW) head, which manipulates the support set to encode causal assumptions during training and inference, encouraging the model to learn representations that rely solely on environment-independent features.
Learning Large Graph Property Prediction via Graph Segment Training
Kaidi Cao (Stanford University), Bryan Perozzi (Google)
Graph Neural NetworkGraph
🎯 What it does: This paper proposes a framework based on Graph Segment Training (GST), which utilizes pre-partitioned graph segments and a historical embedding table to achieve constant memory training for large graph attribute prediction, and mitigates the issue of outdated embeddings through prediction head fine-tuning and old embedding dropout techniques.
Learning Large-Scale MTP$_2$ Gaussian Graphical Models via Bridge-Block Decomposition
Xiwen Wang (Hong Kong University of Science and Technology), Daniel P. Palomar (Hong Kong University of Science and Technology)
OptimizationComputational EfficiencyGraph Neural NetworkGraphTime SeriesAgriculture Related
🎯 What it does: This study investigates the precision matrix learning of large-scale Gaussian graphical models under MTP 2 constraints and proposes an explicit solution to the optimization problem through bridge structure for block decomposition.
Learning Large-scale Neural Fields via Context Pruned Meta-Learning
Jihoon Tack (Korea Advanced Institute of Science and Technology), Jonathan Richard Schwarz (University College London)
Computational EfficiencyMeta LearningNeural Radiance FieldImageVideoMultimodalityAudio
🎯 What it does: An efficient meta-learning framework GradNCP is proposed, which trains large-scale neural fields by online pruning of context points.
Learning Layer-wise Equivariances Automatically using Gradients
Tycho F.A. van der Ouderaa (Imperial College London), Mark van der Wilk (University of Oxford)
ClassificationOptimizationConvolutional Neural NetworkGaussian SplattingImage
🎯 What it does: A framework is proposed that utilizes gradients and Bayesian model selection to automatically learn hierarchical equivariance, capable of discovering symmetry constraints at each layer from training data.
Learning Linear Causal Representations from Interventions under General Nonlinear Mixing
Simon Buchholz (Max Planck Institute for Intelligent Systems), Pradeep Kumar Ravikumar
Representation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: The study investigates how to identify causal representations and latent structures under the conditions of unknown, single-node interventions and arbitrary nonlinear mappings.
Learning List-Level Domain-Invariant Representations for Ranking
Ruicheng Xian (University of Illinois Urbana-Champaign), Michael Bendersky (Google Research)
RetrievalDomain AdaptationAdversarial AttackTransformerText
🎯 What it does: Proposed a list-level domain-invariant representation learning (ListDA) to address the unsupervised domain adaptation problem in ranking tasks, providing a theoretical generalization upper bound and validating its superiority on various retrieval datasets.
Learning Mask-aware CLIP Representations for Zero-Shot Segmentation
Siyu Jiao (Beijing Jiaotong University), Humphrey Shi (Georgia Institute of Technology)
SegmentationKnowledge DistillationTransformerSupervised Fine-TuningContrastive LearningImage
🎯 What it does: This paper proposes the Mask-aware CLIP Fine-Tuning (MAFT) method, which enhances the sensitivity of CLIP to different mask proposals through mask-aware fine-tuning, thereby improving segmentation performance in zero-shot semantic segmentation tasks.
Learning Mixtures of Gaussians Using the DDPM Objective
Kulin Shah (University of Texas at Austin), Adam Klivans (University of Texas at Austin)
Diffusion model
🎯 What it does: This study investigates the efficient learning of parameters for high-dimensional Gaussian mixture models using gradient descent under the objective of the diffusion model DDPM.
Learning Modulated Transformation in GANs
Ceyuan Yang (Shanghai AI Laboratory), Bo Dai (Shanghai AI Laboratory)
GenerationData SynthesisAutonomous DrivingConvolutional Neural NetworkGenerative Adversarial NetworkImageVideo
🎯 What it does: The Modulated Transformation Module (MTM) is proposed, which achieves explicit modeling of geometric deformations in generated images and videos by introducing learnable offsets modulated by latent vectors in convolution.
Learning Motion Refinement for Unsupervised Face Animation
Jiale Tao (University of Electronic Science and Technology of China), Lixin Duan (University of Electronic Science and Technology of China)
GenerationData SynthesisVideo
🎯 What it does: This paper proposes an unsupervised facial animation framework that utilizes a prior coarse motion model (local affine/thin-plate spline) to generate coarse motion flows, and designs a non-prior refinement module based on structural correlation volumes to iteratively refine the motion to capture micro-expressions and detailed movements.
Learning Multi-agent Behaviors from Distributed and Streaming Demonstrations
Shicheng Liu (Pennsylvania State University), Minghui Zhu (Pennsylvania State University)
OptimizationRobotic IntelligenceReinforcement LearningSequential
🎯 What it does: A distributed online inverse reinforcement learning framework MA-BIRDS is proposed, which can learn multi-agent behaviors from distributed real-time demonstrations.
Learning Neural Implicit through Volume Rendering with Attentive Depth Fusion Priors
Pengchong Hu (Wayne State University), Zhizhong Han (Wayne State University)
SegmentationDepth EstimationOptimizationNeural Radiance FieldSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: This paper proposes a method for learning neural implicit representations through voxel rendering, utilizing an attention mechanism to combine Truncated Signed Distance Function (TSDF) with self-learning geometry, thereby enhancing the geometric accuracy of multi-view RGB-D reconstruction.
Learning non-Markovian Decision-Making from State-only Sequences
Aoyang Qin (Tsinghua University), Sirui Xie (University of California, Los Angeles)
Robotic IntelligenceReinforcement Learning from Human FeedbackSequential
🎯 What it does: A generative model called LanMDP based on non-Markov decision processes (nMDP) is proposed to learn implicit actions and make decisions from demonstration sequences that contain only states.
Learning Nonparametric Latent Causal Graphs with Unknown Interventions
Yibo Jiang (University of Chicago), Bryon Aragam (University of Chicago)
Graph Neural NetworkGraph
🎯 What it does: This study investigates how to identify latent causal graphs under unknown latent variable interventions in a non-parametric setting and provides a complete theory of identifiability.
Learning Probabilistic Symmetrization for Architecture Agnostic Equivariance
Jinwoo Kim (Korea Advanced Institute of Science and Technology), Seunghoon Hong (Korea Advanced Institute of Science and Technology)
ClassificationRepresentation LearningTransformerGraph
🎯 What it does: A probabilistic symmetrization framework is proposed, which aligns any base model (such as MLP or Transformer) to be an equivariant function of a given group by learning the distribution of group transformations under input conditions.
Learning Provably Robust Estimators for Inverse Problems via Jittering
Anselm Krainovic (Technical University of Munich), Reinhard Heckel (Technical University of Munich)
RestorationCompressionAdversarial AttackConvolutional Neural NetworkImageMagnetic Resonance Imaging
🎯 What it does: The study achieves robustness in inverse problems by adding Gaussian noise (jittering) during training, proving that under the linear model of subspace denoising, jittering can obtain an optimal robust estimator equivalent to adversarial training, and its effectiveness is validated in tasks such as image denoising, deconvolution, and compressed sensing.
Learning Rate Free Sampling in Constrained Domains
Louis Sharrock (Lancaster University), Christopher Nemeth (Lancaster University)
OptimizationTabular
🎯 What it does: A series of completely learning rate-free particle sampling algorithms are proposed for sampling within constrained domains, along with their theoretical framework and implementation.
Learning Re-sampling Methods with Parameter Attribution for Image Super-resolution
Xiaotong Luo (Xiamen University), Yanyun Qu (Xiamen University)
RestorationSuper ResolutionConvolutional Neural NetworkImage
🎯 What it does: A dual sampling parameter attribution (BSPA) method is proposed, which combines unified sampling and reverse sampling to balance the training data distribution, and uses integrated gradients to filter important parameters, resulting in a more compact super-resolution model.
Learning Regularized Monotone Graphon Mean-Field Games
Fengzhuo Zhang (National University of Singapore), Zhuoran Yang (Yale University)
OptimizationGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: This paper studies the existence and learning problems of λ-regularized graphical mean field games (GMFG), proposing a discrete-time strategy mirror descent algorithm MonoGMFG-PMD and providing a convergence rate.
Learning Reliable Logical Rules with SATNet
Zhaoyu Li (University of Toronto), Xujie Si (Shanghai Jiao Tong University)
OptimizationExplainability and InterpretabilityTabular
🎯 What it does: This study investigates how to decode interpretable logical rules from the weights learned by SATNet, proposing a maximum equation normalization method to convert the weights into readable weighted MaxSAT formulas and verifying their functional equivalence.
Learning Repeatable Speech Embeddings Using An Intra-class Correlation Regularizer
Jianwei Zhang (Arizona State University), Visar Berisha (Arizona State University)
ClassificationRecognitionContrastive LearningAudio
🎯 What it does: This paper introduces the concept of reproducibility from measurement theory into deep embedding learning, evaluates the reproducibility of embeddings using ICC, and further designs an ICC regularizer as a supplement to contrastive loss, promoting the model to learn more reproducible embeddings.
Learning Robust Statistics for Simulation-based Inference under Model Misspecification
Daolang Huang (Aalto University), Samuel Kaski (Aalto University)
Recurrent Neural NetworkAuto EncoderTime Series
🎯 What it does: A robust statistical learning method based on regularization loss is proposed to address model misspecification in simulation-based inference (SBI);
Learning Rule-Induced Subgraph Representations for Inductive Relation Prediction
Tianyu Liu (University of Science and Technology of China), Hanzhu Chen (University of Science and Technology of China)
Representation LearningRecurrent Neural NetworkGraph Neural NetworkGraphBenchmark
🎯 What it does: A single-source edge-level graph neural network (REST) is proposed for inducing subgraph representations, addressing the issue of decreased inference performance caused by the mixing of irrelevant rules within subgraphs.
Learning Sample Difficulty from Pre-trained Models for Reliable Prediction
Peng Cui (Tsinghua University), Jun Zhu (Tsinghua University)
ClassificationOptimizationImage
🎯 What it does: Utilize large-scale pre-trained models to estimate the difficulty of training samples, and use the difficulty information for entropy regularization to improve the accuracy and calibration of downstream models.
Learning Score-based Grasping Primitive for Human-assisting Dexterous Grasping
Tianhao Wu (Peking University), Hao Dong (Peking University)
Robotic IntelligenceReinforcement LearningScore-based ModelPoint Cloud
🎯 What it does: For the assistive grasping task for users with human-like missing arms, a closed-loop control method has been researched and implemented, where a robotic hand palm helps users grasp objects through finger control.
Learning Shared Safety Constraints from Multi-task Demonstrations
Konwoo Kim (Carnegie Mellon University), Steven Wu
Safty and PrivacyRobotic IntelligenceReinforcement LearningSequential
🎯 What it does: This paper proposes an inverse constraint learning method based on multi-task demonstrations to learn shared safety constraints from expert demonstrations.
Learning Space-Time Continuous Latent Neural PDEs from Partially Observed States
Valerii Iakovlev (Aalto University), Harri Lähdesmäki (Aalto University)
TransformerTime SeriesOrdinary Differential Equation
🎯 What it does: A spatiotemporal continuous, mesh-free latent neural PDE model is proposed to learn PDE dynamics from noise and partial observations at irregular spatiotemporal grid points.
Learning the Efficient Frontier
Philippe Chatigny (Riskfuel), Maxime Bergeron (Riskfuel)
OptimizationTransformerTabularTime SeriesFinance Related
🎯 What it does: A Transformer-based SEQ2SEQ model called NeuralEF is proposed for the rapid approximation of the effective frontier problem in high-dimensional financial asset allocation;
Learning threshold neurons via edge of stability
Kwangjun Ahn (Massachusetts Institute of Technology), Yi Zhang (Microsoft Research)
OptimizationImage
🎯 What it does: The paper theoretically analyzes and experimentally investigates the effect of the edge of stability (EoS) on the learning of threshold neurons in two-layer ReLU networks.
Learning Time-Invariant Representations for Individual Neurons from Population Dynamics
Lu Mi (Allen Institute for Brain Science), Uygar Sümbül (Allen Institute for Brain Science)
ClassificationRepresentation LearningTransformerSupervised Fine-TuningMultimodalityBiomedical Data
🎯 What it does: Using a self-supervised neural dynamic model, we learn time-invariant single-cell representations and use these representations to predict the transcriptomic classification of cells.
Learning to Augment Distributions for Out-of-distribution Detection
Qizhou Wang (Hong Kong Baptist University), Bo Han (Hong Kong Baptist University)
Anomaly DetectionOptimizationImage
🎯 What it does: A distribution-augmented OOD learning framework (DAL) is proposed, which reduces the distribution difference between auxiliary OOV and true OOV by searching for the worst OOV distribution within the Wasserstein ball, thereby improving open-world OOD detection performance.
Learning to Compress Prompts with Gist Tokens
Jesse Mu (Stanford University), Noah Goodman (Stanford University)
CompressionComputational EfficiencyMeta LearningTransformerSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: By making slight modifications to the attention mask of the Transformer, the model is trained to insert a small number of 'gist' tokens after the input prompt, allowing the model to compress the entire prompt into these few tokens, which can then be cached to reduce encoding costs.
Learning to Configure Separators in Branch-and-Cut
Sirui Li (Massachusetts Institute of Technology), Cathy Wu (Massachusetts Institute of Technology)
OptimizationGraph Neural NetworkTabularBenchmark
🎯 What it does: This study investigates how to accelerate mixed-integer linear programming (MILP) solvers by learning to select and configure separators within the Branch-and-Cut framework, proposing the 'Separator Configuration' task and providing an end-to-end learning method.
Learning to Discover Skills through Guidance
Hyunseung Kim (KAIST), Jaegul Choo (KAIST)
Robotic IntelligenceReinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: This paper proposes an unsupervised skill discovery algorithm DISCO-DANCE that guides the exploration of unknown states through guide skills.
Learning To Dive In Branch And Bound
Max B. Paulus (ETH Zurich), Andreas Krause (ETH Zurich)
OptimizationGraph Neural NetworkTabular
🎯 What it does: This paper studies how to use generative models based on graph neural networks to predict integer variable assignments, and designs adaptive variable selection and bound tightening rules through linear programming duality, thereby learning branch-and-bound diving heuristics tailored for specific applications.
Learning to Group Auxiliary Datasets for Molecule
Tinglin Huang (Yale University), Zhitao Ying
OptimizationDrug DiscoveryGraph Neural NetworkGraphTabular
🎯 What it does: This paper proposes a method for auxiliary dataset grouping called MolGroup, based on a routing mechanism and dual-layer optimization, aimed at small molecule property prediction tasks. It can automatically identify and select auxiliary datasets that have a high affinity with the target dataset, thereby enhancing model performance.
Learning to Influence Human Behavior with Offline Reinforcement Learning
Joey Hong (University of California Berkeley), Anca Dragan (University of California Berkeley)
Reinforcement LearningSequential
🎯 What it does: Learn the influence strategies of agents when collaborating with real humans based solely on human-human interaction data through offline reinforcement learning (CQL); further adjust human long-term behavior by learning the latent strategy representations of humans.
Learning to Modulate pre-trained Models in RL
Thomas Schmied (Johannes Kepler University), Sepp Hochreiter (Johannes Kepler University)
TransformerReinforcement LearningPrompt EngineeringTabularBenchmark
🎯 What it does: This paper studies how to efficiently fine-tune pre-trained models in reinforcement learning and proposes the L2M method to avoid catastrophic forgetting.
Learning to Parameterize Visual Attributes for Open-set Fine-grained Retrieval
Shijie Wang (Dalian University of Technology), Qi Tian (Huawei Cloud and AI)
RetrievalConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Visual Attribute Parameterization Network (VAPNet) that learns visual attributes from known categories without attribute annotations and parameterizes the retrieval model to enhance the performance of open-set fine-grained retrieval.
Learning to Reason and Memorize with Self-Notes
Jack Lanchantin (Meta AI), Sainbayar Sukhbaatar (Meta AI)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextChain-of-Thought
🎯 What it does: The Self-Notes method is proposed, allowing the model to insert self-notes at any time during the reading process to achieve multi-step reasoning and memory.
Learning to Receive Help: Intervention-Aware Concept Embedding Models
Mateo Espinosa Zarlenga (University of Cambridge), Mateja Jamnik (University of Cambridge)
ClassificationExplainability and InterpretabilityRepresentation LearningImage
🎯 What it does: A new conceptually interpretable model, IntCEM, is studied, which is specifically trained to better utilize intervention information when receiving expert intervention.
Learning to Search Feasible and Infeasible Regions of Routing Problems with Flexible Neural k-Opt
Yining Ma (National University of Singapore), Yeow Meng Chee (National University of Singapore)
OptimizationRecurrent Neural NetworkGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: This paper presents NeuOpt, a learning-search (L2S) solver that can flexibly execute any k-opt, and introduces Guided Infeasible Region Exploration (GIRE) in CVRP to overcome feasibility mask limitations.
Learning to Tokenize for Generative Retrieval
Weiwei Sun (Shandong University), Zhaochun Ren (Leiden University)
GenerationRetrievalTransformerAuto EncoderText
🎯 What it does: This paper proposes an end-to-end learning framework for generative retrieval of document identifiers (docid) called GENRET, replacing traditional rule-based or clustering-based segmentation methods.
Learning Topology-Agnostic EEG Representations with Geometry-Aware Modeling
Ke Yi (South China University of Technology), Dongsheng Li (Microsoft Research Asia)
ClassificationRecognitionDomain AdaptationRepresentation LearningTransformerAuto EncoderTime SeriesBiomedical Data
🎯 What it does: Aiming at EEG data with different electrode configurations, the MMM framework is proposed for self-supervised pre-training, achieving cross-dataset transfer in emotion recognition tasks.
Learning Trajectories are Generalization Indicators
Jingwen Fu (Xi'an Jiaotong University), Nanning Zheng (Xi'an Jiaotong University)
Convolutional Neural NetworkImage
🎯 What it does: This study investigates the relationship between the learning trajectory during the training of deep neural networks (DNN) and generalization error, proposing an analysis from the perspective of the contribution of each update to the generalization error, and provides a new upper bound on generalization error that includes trajectory information.
Learning Transformer Programs
Dan Friedman (Princeton University), Danqi Chen (Princeton University)
TransformerText
🎯 What it does: A Transformer model that can be directly trained and subsequently converted into a readable program—Transformer Programs—is proposed, and its effectiveness is validated across various tasks.
Learning Universal Policies via Text-Guided Video Generation
Yilun Du (Massachusetts Institute of Technology), Pieter Abbeel (University of California Berkeley)
Robotic IntelligenceTransformerReinforcement LearningDiffusion modelVideoTextMultimodality
🎯 What it does: This work proposes learning general strategies through text-guided video generation, transforming action planning into future frame generation using video diffusion models, and then extracting control actions through inverse dynamics networks, thereby achieving compositional generalization and real-world transfer across tasks and environments.
Learning Unseen Modality Interaction
Yunhua Zhang (University of Amsterdam), Cees G. M. Snoek
ClassificationRetrievalRobotic IntelligenceTransformerOptical FlowVideoMultimodalityAudio
🎯 What it does: The problem of 'unseen modality interaction' is proposed, and a multimodal learning framework is presented that can handle missing modality combinations during training and unseen modality combinations during inference.
Learning via Wasserstein-Based High Probability Generalisation Bounds
Paul Viallard (Inria), Benjamin Guedj (Inria)
OptimizationTabularBenchmark
🎯 What it does: Proposes a PAC-Bayes high-probability generalization upper bound based on Wasserstein distance, and designs batch and online learning algorithms based on this.
Learning Visual Prior via Generative Pre-Training
Jinheng Xie (National University of Singapore), Mike Zheng Shou (National University of Singapore)
SegmentationGenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringImageText
🎯 What it does: Discretize visual positions (such as bounding boxes, human keypoints, instance masks) into text sequences that can be processed by Transformers, and use a GPT-2 style decoder-only model for pre-training to learn visual spatial priors; customize sampling of visual layouts through prompts, ultimately used for controlling image synthesis.
Learning with Explanation Constraints
Rattana Pukdee (Carnegie Mellon University), Pradeep Kumar Ravikumar
OptimizationExplainability and InterpretabilitySupervised Fine-TuningText
🎯 What it does: A framework for learning using prior interpretability constraints is proposed, along with theoretical analysis and algorithm implementation.
Learning World Models with Identifiable Factorization
Yu-Ren Liu (Nanjing University), Kun Zhang (Carnegie Mellon University)
Reinforcement LearningAuto EncoderWorld ModelSequential
🎯 What it does: The IFactor framework is proposed, utilizing identifiable block-level factorization to learn world models, and four types of latent states (controllable and reward-related, uncontrollable but reward-related, controllable but reward-unrelated, noise) are used for reinforcement learning.
Learning-to-Rank Meets Language: Boosting Language-Driven Ordering Alignment for Ordinal Classification
Rui Wang (Beijing University of Posts and Telecommunications), Zhaofeng He (China Academy of Sciences)
ClassificationTransformerPrompt EngineeringContrastive LearningImageMultimodality
🎯 What it does: Combining the pre-trained CLIP model with a learned ranking method to enhance ordinal classification performance using language priors.
Leave No Stone Unturned: Mine Extra Knowledge for Imbalanced Facial Expression Recognition
Yuhang Zhang (Beijing University of Posts and Telecommunications), Weihong Deng (Beijing University of Posts and Telecommunications)
ClassificationRecognitionConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: In response to the severely imbalanced data distribution in facial expression recognition, a scheme is proposed to mine minority class information from all training samples and improve the accuracy of minority classes.
Lending Interaction Wings to Recommender Systems with Conversational Agents
Jiarui Jin (Shanghai Jiao Tong University), Jun Wang (University College London)
Recommendation SystemRecurrent Neural NetworkLarge Language ModelReinforcement LearningTabular
🎯 What it does: The CORE framework is proposed, which combines offline-trained recommendation systems with dialogue agents, actively querying attributes or products in conversations through uncertainty minimization.