ICLR 2023 Papers — Page 3
International Conference on Learning Representations · 1573 papers
Capturing the Motion of Every Joint: 3D Human Pose and Shape Estimation with Independent Tokens
Sen Yang (Southeast University), Gang YU
Pose EstimationTransformerImageVideo
🎯 What it does: The paper proposes a Transformer framework based on independent Tokens for estimating 3D human pose and shape from single frames or videos.
CASR: Generating Complex Sequences with Autoregressive Self-Boost Refinement
Hongwei Han (Tsinghua University), Dongmei Zhang (Microsoft Research)
GenerationTransformerSupervised Fine-TuningSequential
🎯 What it does: A self-regressive self-enhancing iterative refinement framework CASR is proposed, using the model's own previous prediction as the input for the next step to gradually improve complex sequence generation.
Causal Balancing for Domain Generalization
Xinyi Wang (University of California), William Yang Wang (University of California)
Domain AdaptationAuto EncoderImageBenchmark
🎯 What it does: A lightweight mini-batch sampling strategy based on causal balancing is proposed for domain generalization tasks. By learning latent covariates and matching balanced scores, unbiased training batches are constructed to suppress spurious correlations.
Causal Confusion and Reward Misidentification in Preference-Based Reward Learning
Jeremy Tien (University of California Berkeley), Daniel S. Brown (University of Utah)
OptimizationReinforcement LearningTabular
🎯 What it does: The system studied the issues of causal confusion and reward misidentification that arise in preference-based reward learning, and evaluated various influencing factors through a series of sensitivity and ablation experiments.
Causal Estimation for Text Data with (Apparent) Overlap Violations
Lin Gui (University of Chicago), Victor Veitch (Google Research)
TransformerSupervised Fine-TuningText
🎯 What it does: This paper studies how to estimate the causal effect of a certain attribute in text data, such as the impact of polite versus rude emails on response time. The authors propose a method to handle causal identification and obtain robust causal estimates in the presence of apparent overlap violations.
Causal Imitation Learning via Inverse Reinforcement Learning
Kangrui Ruan (Columbia University), Elias Bareinboim (Columbia University)
Autonomous DrivingReinforcement LearningGenerative Adversarial NetworkImageTabular
🎯 What it does: This paper proposes achieving imitation learning through Inverse Reinforcement Learning (IRL) in the presence of unobserved confounding variables.
Causal Reasoning in the Presence of Latent Confounders via Neural ADMG Learning
Matthew Ashman (University of Cambridge), Cheng Zhang (Microsoft Research)
Flow-based ModelAuto EncoderTabularSequential
🎯 What it does: A learning framework for ADMG based on neural autoregressive flow (N-ADMG) is proposed, achieving gradient-learnable causal inference and reasoning for causal models with latent confounders and nonlinear structures.
Causal Representation Learning for Instantaneous and Temporal Effects in Interactive Systems
Phillip Lippe (University of Amsterdam), Efstratios Gavves (University of Amsterdam)
Representation LearningFlow-based ModelAuto EncoderTime Series
🎯 What it does: The iCITRIS method is proposed, extending CITRIS to simultaneously learn multidimensional causal variables and causal graphs containing instantaneous causal relationships in time series.
Causality Compensated Attention for Contextual Biased Visual Recognition
Ruyang Liu (Peking University), Ge Li (Peking University)
ClassificationObject DetectionTransformerImage
🎯 What it does: A novel causal compensation attention module called Interventional Dual Attention (IDA) is proposed, which eliminates contextual confounding bias in visual recognition through multi-sampling and attention re-weighting.
Certifiably Robust Policy Learning against Adversarial Multi-Agent Communication
Yanchao Sun (University of Maryland), Furong Huang (University of Maryland)
Adversarial AttackReinforcement Learning
🎯 What it does: A provably robust multi-agent communication defense framework named Ablated Message Ensemble (AME) is proposed, capable of withstanding attacks where up to half of the messages are arbitrarily tampered with during testing.
Certified Defences Against Adversarial Patch Attacks on Semantic Segmentation
Maksym Yatsura (Robert Bosch GmbH), Jan Hendrik Metzen (Robert Bosch GmbH)
SegmentationAdversarial AttackImage
🎯 What it does: A general authentication defense framework called DEMASKED SMOOTHING is proposed to protect semantic segmentation models against adversarial patch attacks.
Certified Training: Small Boxes are All You Need
Mark Niklas Mueller, Martin Vechev (ETH Zurich)
ClassificationImage
🎯 What it does: A certification training method called SABR is proposed, which is based on selecting small and precise sub-regions in the input space for interval propagation;
CFlowNets: Continuous Control with Generative Flow Networks
Yinchuan Li (Huawei Noah's Ark Lab), Jianye HAO
Reinforcement LearningFlow-based Model
🎯 What it does: This paper proposes Continuous Flow Networks (CFlowNets), providing an exploration-based policy learning method for continuous control tasks based on flow matching.
Characteristic Neural Ordinary Differential Equation
Xingzi Xu (Duke University), Vahid Tarokh (Duke University)
ClassificationGenerationFlow-based ModelImageTime SeriesOrdinary Differential Equation
🎯 What it does: This paper proposes Characteristic-Neural Ordinary Differential Equations (C-NODE), which transforms PDEs into ODEs along the characteristic curves of the PDEs, thereby extending the continuous depth framework of NODE to multi-dimensional space while retaining the adjoint training method.
Characterizing intrinsic compositionality in transformers with Tree Projections
Shikhar Murty (Stanford University), Christopher D Manning (Stanford University)
TransformerSequential
🎯 What it does: Proposes an unsupervised, non-parametric tree-structured projection method to measure the tree-structuredness of Transformers in language tasks;
Characterizing the Influence of Graph Elements
Zizhang Chen (Brandeis University), Pengyu Hong (Brandeis University)
Graph Neural NetworkGraph
🎯 What it does: A method based on Influence Functions is proposed to estimate the impact of removing nodes or edges on model parameters and prediction performance in Graph Convolutional Networks (GCN);
Characterizing the spectrum of the NTK via a power series expansion
Michael Murray (University of California Los Angeles), Guido Montufar (Max Planck Institute for Mathematics in the Sciences)
Image
🎯 What it does: In the widthless limit, a power series expansion of the neural tangent kernel (NTK) for fully connected networks of arbitrary depth is derived, and a recursive formula for the coefficients is provided based on Hermite coefficients and network depth.
Chasing All-Round Graph Representation Robustness: Model, Training, and Optimization
Chunhui Zhang (Brandeis University), Chuxu Zhang (Brandeis University)
OptimizationRepresentation LearningAdversarial AttackGraph Neural NetworkMixture of ExpertsGraph
🎯 What it does: A robust training framework for graph neural networks (GNN) based on Mixture of Experts (MoE) is proposed, which enhances the robustness of GNN against graph attacks through the GAME layer, DECOG training strategy, and GRADIV regularization.
Cheap Talk Discovery and Utilization in Multi-Agent Reinforcement Learning
Yat Long Lo (University of Oxford), Shimon Whiteson (University of Oxford)
Reinforcement LearningBenchmark
🎯 What it does: This study investigates the process of discovering and utilizing variable communication channels (cheap talk) in multi-agent reinforcement learning, proposing a two-stage framework based on mutual information maximization and Off-Belief Learning (OBL), and designing a configurable Phone Booth Maze benchmark.
ChiroDiff: Modelling chirographic data with Diffusion Models
Ayan Das (University of Surrey), Yi-Zhe Song (University of Surrey)
GenerationData SynthesisRecurrent Neural NetworkDiffusion modelSequential
🎯 What it does: A non-autoregressive hand-drawn/sketch generation framework called CHIRODIFF based on diffusion models (DDPM) is proposed for modeling vector graphic data in continuous time.
ChordMixer: A Scalable Neural Attention Model for Sequences with Different Length
Ruslan Khalitov (Norwegian University of Science and Technology), Zhirong Yang (Norwegian University of Science and Technology)
ClassificationComputational EfficiencyTransformerTextSequential
🎯 What it does: ChordMixer is proposed, a neural attention model that can be scaled to variable-length sequences; it achieves full receptive field through a parameter-free multi-scale rotation layer and per-channel MLP, maintaining efficient operation on long sequences (up to 1.5M).
Choreographer: Learning and Adapting Skills in Imagination
Pietro Mazzaglia (Ghent University), Sai Rajeswar (ServiceNow Research)
Robotic IntelligenceMeta LearningRecurrent Neural NetworkReinforcement LearningAuto EncoderWorld ModelSequentialBenchmark
🎯 What it does: In this paper, the authors propose Choreographer, an unsupervised reinforcement learning framework based on a world model that can decouple exploration and skill learning in the imagination space, and then use a meta-controller to evaluate and adapt the learned skills in imagination for efficient transfer to downstream tasks.
CircNet: Meshing 3D Point Clouds with Circumcenter Detection
Huan Lei (Australian National University), Hongdong Li (Australian National University)
GenerationData SynthesisGraph Neural NetworkPoint CloudMesh
🎯 What it does: Triangulation of point clouds is achieved by detecting the circumcenters of triangles, using a single-stage neural network to directly output the circumcenter positions within the neighborhood of each point, thereby deriving the corresponding triangles.
CktGNN: Circuit Graph Neural Network for Electronic Design Automation
Zehao Dong (Washington University in St. Louis), Xuan Zhang (Washington University in St. Louis)
OptimizationGraph Neural NetworkGraphBenchmark
🎯 What it does: This paper proposes a circuit graph neural network (CktGNN) that can automatically generate analog circuit topologies and optimize device sizes simultaneously.
CLARE: Conservative Model-Based Reward Learning for Offline Inverse Reinforcement Learning
Sheng Yue (Tsinghua University), Junshan Zhang (University of California)
Reinforcement LearningTabular
🎯 What it does: A conservative model-based reward learning algorithm CLARE for offline inverse reinforcement learning is developed to alleviate reward extrapolation errors.
Classically Approximating Variational Quantum Machine Learning with Random Fourier Features
Jonas Landman (University of Edinburgh), Elham Kashefi (University of Edinburgh)
TabularPhysics Related
🎯 What it does: By analyzing the spectrum of Variational Quantum Circuits (VQC), three classical approximation methods based on Random Fourier Features (RFF) are proposed to approximate their outputs with only knowledge of the VQC architecture.
Clean-image Backdoor: Attacking Multi-label Models with Poisoned Labels Only
Kangjie Chen (Nanyang Technological University), Tianwei Zhang (Nanyang Technological University)
Object DetectionAdversarial AttackImage
🎯 What it does: The first multi-label model backdoor attack method is proposed, which only contaminates training labels without altering the images.
Clifford Neural Layers for PDE Modeling
Johannes Brandstetter (Microsoft Research AI4Science), Jayesh K Gupta
Neural Radiance FieldTime SeriesSequentialPhysics Related
🎯 What it does: Proposed and implemented a neural network layer based on Clifford algebra to accelerate the simulation of PDEs (Navier–Stokes, shallow water equations, and Maxwell's equations).
CLIP-Dissect: Automatic Description of Neuron Representations in Deep Vision Networks
Tuomas Oikarinen (University of California San Diego), Tsui-Wei Weng (University of California San Diego)
Explainability and InterpretabilityRepresentation LearningConvolutional Neural NetworkTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: The CLIP-Dissect technique is proposed, which automatically generates open concept descriptions for hidden neurons in deep visual networks without the need for manual annotation.
CLIP-ViP: Adapting Pre-trained Image-Text Model to Video-Language Alignment
Hongwei Xue (University of Science and Technology of China), Jiebo Luo (Renmin University of China)
RetrievalTransformerVision Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: This study investigates how to further post-pretrain pre-trained image-text models (such as CLIP) on large-scale video-text data and proposes the CLIP-ViP model, which achieves video-text alignment through a video proxy mechanism and full-source cross-modal learning.
CLIPSep: Learning Text-queried Sound Separation with Noisy Unlabeled Videos
Hao-Wen Dong (Sony Group Corporation), Taylor Berg-Kirkpatrick (University of California San Diego)
RecognitionData-Centric LearningContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: A text query-based universal sound separation model CLIPSep is proposed, which is trained solely on unlabeled noisy videos, and its robustness in noisy environments is enhanced through Noise-Invariant Training (NIT).
CO3: Cooperative Unsupervised 3D Representation Learning for Autonomous Driving
Runjian Chen (University of Hong Kong), Ping Luo (University of Hong Kong)
Object DetectionSegmentationAutonomous DrivingRepresentation LearningContrastive LearningPoint Cloud
🎯 What it does: Proposes the CO3 (Cooperative Contrastive Learning and Contextual Shape Prediction) framework, utilizing collaborative point clouds from inside and outside the vehicle for unsupervised 3D representation learning, and transferring to various downstream tasks.
Code Translation with Compiler Representations
Marc Szafraniec (Meta AI), Gabriel Synnaeve (Meta AI)
AI Code AssistantTransformerAuto EncoderText
🎯 What it does: This paper enhances the semantic quality of neural code translation by incorporating LLVM IR information into the code translation model, using three IR-based self-supervised objectives.
CodeBPE: Investigating Subtokenization Options for Large Language Model Pretraining on Source Code
Nadezhda Chirkova (Naver Labs Europe), Sergey Troshin (University of Amsterdam)
GenerationAI Code AssistantTransformerLarge Language ModelText
🎯 What it does: This paper systematically studies the impact of subtokenization schemes on model performance and sequence length in the pre-training of large language models on source code, conducting large-scale experiments based on PLBART and proposing various efficient solutions.
CodeGen: An Open Large Language Model for Code with Multi-Turn Program Synthesis
Erik Nijkamp (Salesforce Research), Caiming Xiong (Salesforce Research)
GenerationAI Code AssistantTransformerLarge Language ModelTextBenchmark
🎯 What it does: A series of large language models CODEGEN, ranging in size from 350M to 16.1B, were trained and released, demonstrating their capabilities in multi-turn program synthesis tasks; simultaneously, a multi-turn programming benchmark MTPB was proposed and made public.
CodeT: Code Generation with Generated Tests
Bei Chen (Microsoft Corporation), Weizhu Chen (Microsoft Corporation)
GenerationAI Code AssistantTransformerLarge Language ModelText
🎯 What it does: This study proposes a framework for automatically generating test cases using the same pre-trained language model and selecting the optimal code solution through bidirectional execution consistency (code solution and test cases).
CogVideo: Large-scale Pretraining for Text-to-Video Generation via Transformers
Wenyi Hong (Tsinghua University), Jie Tang (Tsinghua University)
GenerationData SynthesisTransformerVideoText
🎯 What it does: CogVideo has been constructed and trained, a 9B parameter text-to-video generation Transformer that inherits the pre-trained knowledge of CogView2, utilizing multi-frame rate training and a dual-channel attention mechanism to achieve large-scale text-to-video generation.
Collaborative Pure Exploration in Kernel Bandit
Yihan Du (Tsinghua University), Longbo Huang (Tsinghua University)
🎯 What it does: A collaborative pure exploration (CoPE-KB) model is proposed, allowing multiple agents to cooperate in finding the optimal arm across different but related tasks, with algorithms CoKernelFC and CoKernelFB provided for fixed confidence and fixed budget objectives.
Combating Exacerbated Heterogeneity for Robust Models in Federated Learning
Jianing Zhu (Hong Kong Baptist University), Bo Han (Hong Kong Baptist University)
Federated LearningAdversarial AttackImage
🎯 What it does: This study addresses the issue of robustness decline caused by the combination of adversarial training and federated learning, proposing the Slack Federated Adversarial Training (SFAT) framework to alleviate the exacerbated heterogeneity due to adversarial samples.
Combinatorial Pure Exploration of Causal Bandits
Nuoya Xiong (Tsinghua University), Wei Chen (Microsoft Research)
🎯 What it does: A pure exploration (PAC) algorithm is proposed for combinatorial intervention sets in the case of known causal graphs but unknown distributions, implemented for both binary generalized linear models (BGLM) and general graphs with latent variables.
Combinatorial-Probabilistic Trade-Off: P-Values of Community Properties Test in the Stochastic Block Models
Shuting Shen (Harvard University), Junwei Lu (Harvard University)
Graph
🎯 What it does: A framework is proposed for inferring the community combinatorial properties of random block models, aimed at testing whether specific community properties hold and providing p-values to quantify uncertainty.
Competitive Physics Informed Networks
Qi Zeng (Georgia Institute of Technology), Florian Tobias Schaefer
Generative Adversarial NetworkPhysics Related
🎯 What it does: This paper proposes a Competitive Physics-Informed Neural Network (CPINN) that solves partial differential equations (PDEs) through adversarial training between a discriminator network and a PINN, addressing the issue of traditional PINNs struggling to achieve high accuracy.
Complexity-Based Prompting for Multi-step Reasoning
Yao Fu (University of Edinburgh), Tushar Khot (Allen Institute for AI)
Large Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: This paper studies prompt strategies for multi-step reasoning, proposing example selection and decoding based on reasoning complexity.
Composing Ensembles of Pre-trained Models via Iterative Consensus
Shuang Li (Massachusetts Institute of Technology), Igor Mordatch (Google)
GenerationRobotic IntelligenceTransformerLarge Language ModelDiffusion modelImageVideoMultimodality
🎯 What it does: This paper proposes a unified framework that utilizes the generator of pre-trained models and multiple evaluators to accomplish zero-shot multimodal tasks through iterative consensus.
Composing Task Knowledge With Modular Successor Feature Approximators
Wilka Torrico Carvalho (University of Michigan), Satinder Singh (University of Michigan)
TransformerReinforcement LearningMultimodality
🎯 What it does: A modular Successor Feature Approximator (MSFA) is proposed, which achieves zero-shot composition of task knowledge by learning predictable state features and successor features through modular learning.
Composite Slice Transformer: An Efficient Transformer with Composition of Multi-Scale Multi-Range Attentions
Mingu Lee (Qualcomm AI Research), Christopher Lott (Qualcomm AI Research)
Computational EfficiencyTransformerText
🎯 What it does: This paper proposes the Composite Slice Transformer (CST), which enhances the efficiency and modeling capability of the Transformer by chunking slice sequences and concatenating local fine-grained and global coarse-grained attention.
Compositional Law Parsing with Latent Random Functions
Fan Shi (Fudan University), Xiangyang Xue (Fudan University)
Explainability and InterpretabilityRepresentation LearningImage
🎯 What it does: A concept-level rule parsing framework (CLAP) based on deep latent variable models is proposed, which automatically parses natural or artificial rules in scenes by decomposing images into independent concepts and learning latent stochastic functions for each concept, thereby enabling predictions of future states and scene reconstruction.
Compositional Prompt Tuning with Motion Cues for Open-vocabulary Video Relation Detection
Kaifeng Gao (Zhejiang University), Qianru Sun (Singapore Management University)
RecognitionObject DetectionKnowledge DistillationTransformerPrompt EngineeringVision Language ModelVideoMultimodality
🎯 What it does: This paper proposes the Open-VidVRD task for open vocabulary video visual relationship detection and designs a combination-based prompt tuning framework called RePro, which can be trained on a limited set of benchmark categories and generalized to unseen object and predicate categories.
Compositional Semantic Parsing with Large Language Models
Andrew Drozdov (Google Research), Denny Zhou (Google Research)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposes a dynamic least-to-most prompting strategy that achieves problem decomposition in large language models through syntactic parsing, enhancing the combinatorial generalization of semantic parsing.
Compositional Task Representations for Large Language Models
NAN SHAO, Zhilin Yang (Tsinghua University)
ClassificationGenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a task representation method that does not rely on prompts—Compositional Task Representations (CTR). It learns a discrete, composable code table through multi-task training to achieve cross-task generalization in the absence of labels or with a small number of labels.
Compositionality with Variation Reliably Emerges in Neural Networks
Henry Conklin (University of Edinburgh), Kenny Smith (University of Edinburgh)
Reinforcement LearningText
🎯 What it does: The study investigates whether the language that emerges in neural networks during communication tasks possesses compositionality, and points out its often underestimated variability in natural language.
Compressing multidimensional weather and climate data into neural networks
Langwen Huang (ETH Zurich), Torsten Hoefler (ETH Zurich)
CompressionTime Series
🎯 What it does: A method for compressing multi-dimensional weather and climate data using coordinate-based neural networks is proposed.
Computational Language Acquisition with Theory of Mind
Andy Liu (Harvey Mudd College), Graham Neubig (Carnegie Mellon University)
GenerationConvolutional Neural NetworkRecurrent Neural NetworkLarge Language ModelReinforcement LearningImageText
🎯 What it does: This study constructs language learners with Theory of Mind (ToM) capabilities in an image referencing game environment, using an internal 'listener' model to reorder candidate sentences, thereby achieving more persuasive and information-rich language expression.
Computing all Optimal Partial Transports
Abhijeet Phatak (Walmart Global Tech), Kaiyi Zhang (Virginia Tech)
Anomaly DetectionOptimizationImageTabular
🎯 What it does: This paper proposes an algorithm for computing the complete optimal transport (OT-profile) of discrete distributions and utilizes the first derivative of the OT-profile to achieve anomaly detection and class prior estimation in unsupervised learning.
Concept Gradient: Concept-based Interpretation Without Linear Assumption
Andrew Bai (University of California, Los Angeles), Cho-Jui Hsieh (University of California, Los Angeles)
Explainability and InterpretabilityBiomedical Data
🎯 What it does: The study proposes the Concept Gradients (CG) method, which provides concept-level explanations for black-box models without the need for linear assumptions.
Concept-level Debugging of Part-Prototype Networks
Andrea Bontempelli (University of Trento), Andrea Passerini (University of Trento)
OptimizationExplainability and InterpretabilitySupervised Fine-TuningImage
🎯 What it does: Proposes ProtoPDebug, an interactive debugger based on conceptual levels, to correct erroneous predictions in ProtoPNets caused by confounding factors; it allows human supervisors to label which parts of the prototypes (part-prototypes) are noise or useful, and then fine-tunes the model using two new penalty terms (forgetting and remembering);
Conditional Antibody Design as 3D Equivariant Graph Translation
Xiangzhe Kong (Tsinghua University), Yang Liu (Tsinghua University)
Drug DiscoveryProtein Structure PredictionGraph Neural NetworkGraph
🎯 What it does: A multi-channel equivariant attention network (MEAN) is proposed, modeling antibody design as conditional 3D graph translation, jointly predicting C1R sequences and structures.
Conditional Positional Encodings for Vision Transformers
Xiangxiang Chu (Meituan Inc), Chunhua Shen (Zhejiang University)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: Proposed and implemented Conditional Position Encoding (CPE), which generates dynamically adjustable position encodings through local neighborhoods, allowing Transformer models like ViT to maintain translational equivalence and improve accuracy across different input sizes.
Confidence Estimation Using Unlabeled Data
Chen Li (Stony Brook University), Chao Chen (Stony Brook University)
ClassificationSegmentationImage
🎯 What it does: A confidence estimation method based on training consistency (consistency ranking loss) is proposed, suitable for semi-supervised learning scenarios, utilizing the prediction consistency of unlabeled samples as an approximation of unlabeled confidence and calibrating the model's softmax output.
Confidence-Based Feature Imputation for Graphs with Partially Known Features
Daeho Um (Seoul National University), Jin young Choi
Graph Neural NetworkGraph
🎯 What it does: In response to the challenge of missing node features in graph learning tasks, this paper proposes a pseudo-confidence-based feature imputation method (PCFI) to recover node features under high missing rates for downstream tasks.
Confidence-Conditioned Value Functions for Offline Reinforcement Learning
Joey Hong (University of California), Sergey Levine (University of California)
Reinforcement LearningTabular
🎯 What it does: This paper proposes an offline reinforcement learning framework called Confidence-Conditioned Value Learning (CCVL), which can learn value functions under different levels of conservativeness (confidence) during training and dynamically adjust the conservativeness based on online observations during evaluation, thereby achieving adaptive policies.
Confidential-PROFITT: Confidential PROof of FaIr Training of Trees
Ali Shahin Shamsabadi (Alan Turing Institute), Adrian Weller (University of Cambridge)
OptimizationSafty and PrivacyReinforcement LearningTabular
🎯 What it does: This paper proposes Confidential-PROFITT, which can prove that the decision tree training process complies with fairness constraints using zero-knowledge proofs without disclosing training data or models.
Conservative Bayesian Model-Based Value Expansion for Offline Policy Optimization
Jihwan Jeong (University of Toronto), Scott Sanner (University of Toronto)
OptimizationReinforcement LearningTabular
🎯 What it does: A Conservative Bayesian Model-Based Value Expansion (CBOP) method is proposed for policy evaluation and optimization in offline reinforcement learning.
Consolidator: Mergable Adapter with Group Connections for Visual Adaptation
Tianxiang Hao (Tsinghua University), Guiguang Ding (Tsinghua University)
ClassificationObject DetectionSegmentationTransformerSupervised Fine-TuningImage
🎯 What it does: A mergeable Grouped Convolution (GC) module, named consolidator, is proposed, which efficiently transfers visual downstream tasks by training only a small number of mergeable branches while keeping the original visual Transformer backbone frozen.
Constraining Representations Yields Models That Know What They Don't Know
Joao Monteiro (ServiceNow Research), David Vazquez (ServiceNow Research)
ClassificationExplainability and InterpretabilityAdversarial AttackConvolutional Neural NetworkTransformerImageText
🎯 What it does: A total activation classifier (TAC) is designed to constrain the activation patterns by using pre-defined binary class codes in the internal layers of the network, enabling the discrimination of model prediction confidence.
Constructive TT-representation of the tensors given as index interaction functions with applications
Gleb Ryzhakov (Skolkovo Institute of Science and Technology), Ivan Oseledets (Skolkovo Institute of Science and Technology)
🎯 What it does: A fast method for directly constructing TT decomposition based on known analytical forms of tensors is proposed;
Context-enriched molecule representations improve few-shot drug discovery
Johannes Schimunek (Johannes Kepler University Linz), Günter Klambauer (Merck Healthcare)
Drug DiscoveryTabularBiomedical DataRetrieval-Augmented Generation
🎯 What it does: A context-enhanced model MHNfs based on modern Hopfield networks is proposed for few-shot drug discovery;
Contextual bandits with concave rewards, and an application to fair ranking
Virginie Do (PSL University), Nicolas Usunier (Meta AI)
Recommendation SystemOptimizationReinforcement LearningTabular
🎯 What it does: This paper proposes a general Contextual Bandit with Multiple Objectives (CBCR) framework and presents an algorithm that achieves asymptotic zero regret without restrictions on the policy space.
Contextual Convolutional Networks
Shuxian Liang (Zhejiang University), Xian-Sheng Hua (Zhejiang University)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkImageVideo
🎯 What it does: A novel CNN backbone network named Contextual Convolutional Network is proposed, which dynamically adjusts the convolutional kernel weights and sampling positions by using the top-k categories from the previous layer as contextual priors, thereby achieving context-aware feature extraction.
Contextual Image Masking Modeling via Synergized Contrasting without View Augmentation for Faster and Better Visual Pretraining
Shaofeng Zhang (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
Object DetectionSegmentationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: A framework called ccMIM is proposed, which combines context mask modeling with contrastive learning in visual pre-training. It actively selects semantically rich patches for masking through attention importance sampling and aligns the global tokens of masked and unmasked patches under a single view using contrastive loss, thereby enhancing the learning difficulty and convergence speed of MIM.
Continual evaluation for lifelong learning: Identifying the stability gap
Matthias De Lange (KU Leuven), Tinne Tuytelaars (KU Leuven)
ClassificationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: Proposed a continuous evaluation framework and identified a stability gap in continuous learning methods.
Continual Pre-training of Language Models
Zixuan Ke (University of Illinois at Chicago), Bing Liu (University of Illinois at Chicago)
Domain AdaptationTransformerLarge Language ModelContrastive LearningTextBiomedical Data
🎯 What it does: The DAS method is proposed to achieve continual domain adaptation pre-training (continual DAP-training) to enhance the performance of language models in new domains while preventing catastrophic forgetting.
Continual Transformers: Redundancy-Free Attention for Online Inference
Lukas Hedegaard (Aarhus University), Alexandros Iosifidis (Aarhus University)
Object DetectionComputational EfficiencyTransformerVideoTime SeriesAudio
🎯 What it does: A Transformer architecture capable of online inference without redundancy and step-by-step reasoning—Continual Transformers—is proposed, introducing a new continuous Scaled Dot-Product Attention (Retroactive and Single-Output) in the Transformer Encoder, while adapting time series with Recycling Positional Encoding.
Continual Unsupervised Disentangling of Self-Organizing Representations
Zhiyuan Li (Rochester Institute of Technology), Linwei Wang (Stanford University)
GenerationRepresentation LearningSpiking Neural NetworkAuto EncoderImage
🎯 What it does: A continuous unsupervised learning framework called CUDOS is proposed, which utilizes a self-organizing SOM mixture model and a spike-and-slab prior VAE to automatically mine the active semantic factors of each data environment and accumulate the relationships between them, achieving continuous semantic factor separation and reuse.
Continuized Acceleration for Quasar Convex Functions in Non-Convex Optimization
Jun-Kun Wang (Yale University), Andre Wibisono (Yale University)
OptimizationTabularStochastic Differential Equation
🎯 What it does: This paper proposes a new acceleration algorithm that utilizes a continuous Nesterov acceleration method to minimize non-convex functions that satisfy quasar convexity, addressing the issue of requiring multiple gradient evaluations in previous methods.
Continuous PDE Dynamics Forecasting with Implicit Neural Representations
Yuan Yin (Sorbonne Université), patrick gallinari
Time SeriesPhysics RelatedOrdinary Differential Equation
🎯 What it does: Proposed the DINO model, which combines implicit neural representation (INR) with neural ODE to achieve continuous spatiotemporal prediction of PDE dynamics;
Continuous pseudo-labeling from the start
Dan Berrebbi (Carnegie Mellon University), Tatiana Likhomanenko (Apple)
RecognitionTransformerSupervised Fine-TuningAudio
🎯 What it does: This paper proposes a self-training framework that allows for continuous pseudo-labeling from the beginning of the training phase, eliminating the traditional pre-training step.
Continuous-Discrete Convolution for Geometry-Sequence Modeling in Proteins
Hehe Fan (Zhejiang University), Mohan Kankanhalli (National University of Singapore)
Protein Structure PredictionConvolutional Neural NetworkBiomedical Data
🎯 What it does: A continuous-discrete convolution (CDConv) network is proposed to simultaneously process protein sequences (1D) and geometric (3D) information for protein structure and function prediction.
Continuous-time identification of dynamic state-space models by deep subspace encoding
Gerben I. Beintema (Eindhoven University of Technology), Roland Tóth (Eindhoven University of Technology)
Time SeriesBenchmarkOrdinary Differential Equation
🎯 What it does: A new estimation method called the Subspace Encoder Method (SUBNET) is proposed for identifying dynamic state space models, particularly continuous-time nonlinear state space models, addressing issues related to external inputs, measurement noise, and latent states in experimental settings.
ContraNorm: A Contrastive Learning Perspective on Oversmoothing and Beyond
Xiaojun Guo (Peking University), Yisen Wang (Peking University)
OptimizationRepresentation LearningGraph Neural NetworkTransformerContrastive LearningImageTextGraph
🎯 What it does: A normalization layer called ContraNorm based on contrastive learning uniformity loss is proposed to alleviate the issues of over-smoothing and dimensional collapse in GNNs and Transformers.
Contrastive Alignment of Vision to Language Through Parameter-Efficient Transfer Learning
Zaid Khan (Northeastern University), Yun Fu (Northeastern University)
RetrievalRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A parameter-efficient alignment method for contrastive vision-language models (LilT) is proposed, which only requires updating a small number of parameters from the original pre-trained vision and language models to achieve CLIP performance comparable to full model training.
Contrastive Audio-Visual Masked Autoencoder
Yuan Gong (Massachusetts Institute of Technology), James R. Glass (Massachusetts Institute of Technology)
ClassificationRetrievalRepresentation LearningTransformerAuto EncoderContrastive LearningVideoMultimodalityAudio
🎯 What it does: This paper proposes CAV-MAE, a self-supervised pre-training model for audio-visual multimodal learning that simultaneously utilizes masked autoencoding and contrastive learning.
Contrastive Corpus Attribution for Explaining Representations
Chris Lin (University of Washington), Su-In Lee (University of Washington)
Explainability and InterpretabilityRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningImageTextMultimodality
🎯 What it does: The COCOA (Contrastive Corpus Attribution) method is proposed to explain the representation vectors of unsupervised models by evaluating the importance of input features through a comparison between the corpus and a foil set.
Contrastive Learning Can Find An Optimal Basis For Approximately View-Invariant Functions
Daniel D. Johnson (University of Toronto), Chris J. Maddison (University of Toronto)
OptimizationRepresentation LearningContrastive LearningImage
🎯 What it does: Proposes viewing contrastive learning as kernel learning of an approximate positive-pair kernel and proves that performing Kernel PCA on this kernel yields the optimal linear predictor;
Contrastive Learning for Unsupervised Domain Adaptation of Time Series
Yilmazcan Ozyurt (ETH Zurich), Ce Zhang (ETH Zurich)
Domain AdaptationRecurrent Neural NetworkContrastive LearningTime SeriesBiomedical Data
🎯 What it does: A contrastive learning-based unsupervised domain adaptation framework CLUDA is proposed, aimed at transferring labeled learning from the source domain to multivariate time series data in the target domain.
Contrastive Meta-Learning for Partially Observable Few-Shot Learning
Adam Jelley (University of Edinburgh), Sam Devlin (Microsoft Research)
Representation LearningMeta LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: By using POEM, some observable multi-view data is mapped to an uncertainty representation, and a unified representation is obtained through a product expert mechanism to achieve representation learning in few-shot learning.
Copy is All You Need
Tian Lan (Tencent AI Lab), Xian-Ling Mao (Beijing Institute of Technology)
GenerationRetrievalTransformerTextRetrieval-Augmented Generation
🎯 What it does: The text generation process is changed to gradually copy phrases from a large-scale text collection, rather than predicting word by word from a fixed vocabulary.
Correlative Information Maximization Based Biologically Plausible Neural Networks for Correlated Source Separation
Bariscan Bozkurt (Koc University), Alper Tunga Erdogan
Recurrent Neural NetworkVideo
🎯 What it does: A biologically plausible neural network framework based on maximizing relevant information, CorInfoMax, is proposed for unsupervised separation of relevant sources.
Corrupted Image Modeling for Self-Supervised Visual Pre-Training
Yuxin Fang (Huazhong University of Science and Technology), Furu Wei (Microsoft Research)
ClassificationSegmentationRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: This paper proposes a self-supervised visual pre-training framework named CIM (Corrupted Image Modeling), which utilizes a trainable BEiT generator to corrupt input images and trains an enhancer (ViT or CNN) to learn visual representations.
CoRTX: Contrastive Framework for Real-time Explanation
Yu-Neng Chuang (Rice University), Xia Hu (Rice University)
Explainability and InterpretabilityContrastive LearningImageTabular
🎯 What it does: A real-time explanation framework CoRTX based on contrastive learning is developed, utilizing unlabeled potential explanation vectors and fine-tuning the explanation head with a minimal number of explanation labels.
Coupled Multiwavelet Operator Learning for Coupled Differential Equations
Xiongye Xiao (University of Southern California), Paul Bogdan (University of Southern California)
Time SeriesPhysics Related
🎯 What it does: A Coupled Multi-Wavelet Neural Operator (CMWNO) is proposed, which achieves unsupervised learning and solving of coupled partial differential equations by decoupling the coupling integral kernel in the multi-wavelet domain and embedding mutual information during the reconstruction process.
Coverage-centric Coreset Selection for High Pruning Rates
Haizhong Zheng (University of Michigan), Atul Prakash (University of Michigan)
OptimizationConvolutional Neural NetworkImage
🎯 What it does: A coverage-centered single-core subset selection method CCS is proposed, aimed at maintaining model performance under high pruning rates;
CrAM: A Compression-Aware Minimizer
Alexandra Peste (Institute of Science and Technology Austria), Dan Alistarh (Neural Magic)
CompressionOptimizationConvolutional Neural NetworkTransformerSupervised Fine-TuningImageText
🎯 What it does: A compressed sensing minimizer CrAM was trained, capable of generating models that are easy to compress in one go without a drop in accuracy during training.
Critic Sequential Monte Carlo
Vasileios Lioutas (Inverted AI), Adam Scibior (Inverted AI)
Autonomous DrivingReinforcement LearningSequential
🎯 What it does: A planning algorithm based on Sequential Monte Carlo (SMC) called CriticSMC is proposed, which utilizes the learned soft Q function as a heuristic factor to guide particles in efficiently exploring environments with sparse hard constraints.
CROM: Continuous Reduced-Order Modeling of PDEs Using Implicit Neural Representations
Peter Yichen Chen (Columbia University), Eitan Grinspun (University of Toronto)
OptimizationComputational EfficiencyAuto EncoderPoint CloudMeshPhysics Related
🎯 What it does: A continuous low-rank modeling framework called CROM is proposed, which is discretization-independent and uses implicit neural fields to represent the continuous field of PDE solutions and evolve in a low-dimensional latent space, significantly improving solution speed and memory efficiency.
Cross-Layer Retrospective Retrieving via Layer Attention
Yanwen Fang (University of Hong Kong), Guodong Li (University of Hong Kong)
ClassificationObject DetectionSegmentationRepresentation LearningConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a cross-layer attention mechanism called MRLA (and its lightweight version MRLA-light), which enhances inter-layer interaction by recursively allowing the current layer to query features from all previous layers, thereby improving the representational capability of deep networks.
Cross-Level Distillation and Feature Denoising for Cross-Domain Few-Shot Classification
Hao ZHENG (Tokyo Institute of Technology), Asako Kanezaki (Tokyo Institute of Technology)
ClassificationDomain AdaptationKnowledge DistillationConvolutional Neural NetworkContrastive LearningImageBenchmark
🎯 What it does: In the cross-domain few-shot classification scenario, a method is proposed that utilizes a small number of unlabeled target domain samples, employing cross-layer knowledge distillation and feature denoising to achieve effective transfer based on source domain knowledge.
Crossformer: Transformer Utilizing Cross-Dimension Dependency for Multivariate Time Series Forecasting
Yunhao Zhang (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
TransformerTime Series
🎯 What it does: Proposes Crossformer, a Transformer model specifically designed for multivariate time series forecasting, which explicitly utilizes cross-dimensional dependencies;
CUDA: Curriculum of Data Augmentation for Long-tailed Recognition
Sumyeong Ahn (KAIST AI), Se-Young Yun (KAIST AI)
ClassificationRecognitionData-Centric LearningReinforcement LearningImage
🎯 What it does: Proposes CUDA, an adaptive class-level data augmentation curriculum to alleviate class imbalance in long-tail classification.
Curriculum-based Co-design of Morphology and Control of Voxel-based Soft Robots
Yuxing Wang (Tsinghua University), Xueqian Wang (Tsinghua University)
OptimizationRobotic IntelligenceTransformerReinforcement LearningSequential
🎯 What it does: A course learning-based co-design method for Voxel soft-bodied robots and control, named CuCo, is proposed, which gradually expands the design space while simultaneously learning design and control strategies.