ICML 2026 Papers — Page 12
International Conference on Machine Learning · 6554 papers
Continuous-Time Piecewise-Linear Recurrent Neural Networks
Alena Brändle, Daniel Durstewitz (Central Institute of Mental Health)
OptimizationExplainability and InterpretabilityComputational EfficiencyRecurrent Neural NetworkAuto EncoderContrastive LearningTime SeriesSequentialBiomedical DataPhysics RelatedStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Propose a continuous-time piecewise linear recurrent neural network (cPLRNN) and an analytical solving algorithm without numerical integration, supporting exact inference and prediction at any time point;
CONTINUUM: Restoring the Contiguous Tensor Abstraction Efficiently for Dynamic AI Workloads via Hardware Virtualization
Yangyu Zhang (Chinese Academy of Sciences), Jiacheng Zhao (Chinese Academy of Sciences)
OptimizationComputational EfficiencyLarge Language ModelPrompt EngineeringTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Developed CONTINUUM, a middleware that utilizes GPU hardware virtual memory to achieve tensor virtualization, restoring the contiguous tensor abstraction and providing an elastic tensor interface.
Contractive Anchor Resolvent Diffusion for Incomplete Multi-View Clustering
Tongzheng Zhao (Shanxi University), Liang Du (Shanxi University)
OptimizationComputational EfficiencyRepresentation LearningGraph Neural NetworkDiffusion modelScore-based ModelAuto EncoderContrastive LearningMultimodalityGraph
🎯 What it does: Propose a framework named CARD, which handles incomplete multi-view clustering from a spectral filtering perspective, performing high-order resolvent diffusion directly on the observed relational graph without explicitly imputing missing views.
Contrastive Diffusion Alignment: Learning Structured Latents for Controllable Generation
Ruchi Sandilya (Weill Cornell Medicine), Logan Grosenick (Weill Cornell Medicine)
GenerationRepresentation LearningDiffusion modelAuto EncoderContrastive LearningImageVideoTime SeriesBiomedical Data
🎯 What it does: This paper proposes the ConDA framework, which uses contrastive learning to map the high-dimensional latent space of diffusion models into a low-dimensional, dynamically aligned embedding space. Nonlinear trajectory editing is performed in this space, and then a kNN decoder is used to return to the original latent space for rendering, achieving controllable generation.
Contrastive Flow Map Matching
Junyu Zhang (Sungkyunkwan University), Eunbyung Park (Yonsei University)
GenerationData SynthesisTransformerDiffusion modelScore-based ModelFlow-based ModelContrastive LearningOptical FlowImage
🎯 What it does: Propose the Contrastive Flow Map Matching (CFMM) framework, which improves the training of flow map matching models by better aligning the training objectives with the actual inference process.
Contrastive Geometric Learning Unlocks Unified Structure- and Ligand-Based Drug Design
Lisa Schneckenreiter (Johannes Kepler University), Günter Klambauer (Johannes Kepler University)
Drug DiscoveryProtein Structure PredictionGraph Neural NetworkTransformerDiffusion modelGenerative Adversarial NetworkContrastive LearningGraphBiomedical Data
🎯 What it does: Designed a unified contrastive geometric learning model, ConGLUDe, which can be jointly trained on structure-based and ligand-based data, achieving multi-task functions such as pocket prediction, virtual screening, target capture, and ligand-conditioned pocket selection.
Contrastive Order Learning: A General Framework for Ordinal Regression
Chaewon Lee (Korea University), Chang-Su Kim (Korea University)
Image TranslationRestorationRecommendation SystemAnomaly DetectionOptimizationFederated LearningComputational EfficiencyKnowledge DistillationRepresentation LearningAdversarial AttackHyperparameter SearchData-Centric LearningRobotic IntelligenceMeta LearningDrug DiscoveryAI Code AssistantReinforcement Learning from Human FeedbackNeural Architecture SearchProtein Structure PredictionTransformerAuto EncoderGenerative Adversarial NetworkContrastive LearningImageVideo
🎯 What it does: Proposed a general framework called ConOrd that combines contrastive learning with ordinal learning, performing ordinal contrastive learning on all sample pairs using soft weights;
Contrastive Reasoning Alignment: Reinforcement Learning from Hidden Representations
Haozheng Luo (Northwestern University), Yan Chen (Northwestern University)
Safty and PrivacyReinforcement Learning from Human FeedbackTransformerReinforcement LearningContrastive LearningTextChain-of-Thought
🎯 What it does: This paper proposes a latent space alignment framework called CRAFT based on contrastive learning and reinforcement learning, which uses the hidden representations of large inference models to enhance their security against jailbreak attacks.
Contrastive Representation Regularization for Vision-Language-Action Models
Taeyoung Kim (KAIST), Jinwoo Shin (KAIST)
Representation LearningRobotic IntelligenceTransformerVision-Language-Action ModelFlow-based ModelContrastive LearningImageTextMultimodality
🎯 What it does: Propose a contrastive regularization based on the robot's body state (RS-CL), which directly performs contrastive learning on VLM representations during VLA model training, making them more aligned with the robot's joint/end-effector position and posture, thereby improving action prediction performance.
Contrastive Spectral Rectification: Test-Time Defense towards Zero-shot Adversarial Robustness of CLIP
Sen Nie (Chinese Academy of Sciences), Xilin CHEN
ClassificationRestorationRepresentation LearningAdversarial AttackTransformerContrastive LearningGaussian SplattingImageTextMultimodality
🎯 What it does: This paper studies the frequency domain vulnerability of CLIP on adversarial samples, and proposes a test-time defense method called CSR based on this;
Contrastive Symbolic Regression: Aligned Representations, Adaptive Prediction, and Diverse Ensembles
Hengzhe Zhang (Victoria University of Wellington), Mengjie Zhang (Victoria University of Wellington)
OptimizationComputational EfficiencyRepresentation LearningData-Centric LearningContrastive LearningTabular
🎯 What it does: Proposed Contrastive Symbolic Regression (CSR), which combines evolutionary feature construction with contrastive learning to generate representations consistent with the geometry of the target space, and improves prediction performance using KNN regression and ensemble methods.
Contrastive Weak-to-Strong Generalization
Houcheng Jiang (University Of Science And Technology Of China), Yang Deng (Singapore Management University)
Knowledge DistillationRepresentation LearningReinforcement Learning from Human FeedbackTransformerSupervised Fine-TuningReinforcement LearningContrastive LearningText
🎯 What it does: Proposes the Contrastive Weak-to-Strong Generalization (ConG) framework, which utilizes contrastive decoding to generate high-quality samples from a weak model for training a stronger model, achieving generalization from a weak model to a strong model.
ContrastiveCFG: Guiding Diffusion Sampling by Contrasting Positive and Negative Concepts
Jinho Chang (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisPrompt EngineeringDiffusion modelScore-based ModelContrastive LearningImage
🎯 What it does: Propose ContrastiveCFG (CCFG), a sampling method that utilizes contrastive loss to guide diffusion model sampling, achieving control over both positive and negative concepts.
Contribution Weights: A Geometrical Analysis of Self-Attention Transformers
Harry Jake Cunningham (University College London), Nicola Muca Cirone (Cartesia.AI)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Proposed and evaluated a contribution weight metric that considers both the magnitude and direction of value vectors, as an alternative to traditional attention weights for explaining the importance of tokens in large language models.
Control Consistency Losses for Diffusion Bridges
Samuel Howard (University of Oxford), Jakiw Pidstrigach (University of Oxford)
GenerationData SynthesisOptimizationComputational EfficiencyDiffusion modelScore-based ModelTabularTime SeriesBiomedical DataPhysics RelatedStochastic Differential Equation
🎯 What it does: Proposed a loss function based on control self-consistency to online learn and simulate diffusion bridges without requiring gradients of simulated trajectories during training.
Controllable and Explainable Personality Sliders for LLMs at Inference Time
Florian Hoppe (Technical University of Munich), Mark Huasong Meng (Technical University of Munich)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Propose a method to achieve controllable and interpretable modulation of the Big Five personality traits of large language models (LLMs) during inference by leveraging activation vectors
Controllable Molecule Generation via Sparse Representation Editing: An Interpretability-Driven Perspective
Zhuoran Li (Hong Kong Polytechnic University), Wanyu Lin (Hong Kong Polytechnic University)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningDrug DiscoveryTransformerLarge Language ModelAuto EncoderContrastive LearningTextGraphBiomedical Data
🎯 What it does: Constructed the SpaRE framework, utilizing sparse representation editing to achieve fine-grained controllability over molecules generated by LLMs, supporting both local (atom/functional group) and global (property) control.
Controlled Collaboration Geometry for Personalized Federated Learning
Hongbo Yin (University of Electronic Science and Technology of China), Yan Zhang (University of Electronic Science and Technology of China)
Federated LearningContrastive LearningImageTextTabular
🎯 What it does: Proposed a personalized federated learning framework called pFedCCG based on control collaborative geometry, which avoids consistency collapse and self-clustering through static similarity templates, objective alignment projection, and collaborative intensity scheduling.
Controlled Dynamics Attractor Transformer
Cheng Zhang (Xi'an Jiaotong University), Qinghua Zheng (Xi'an Jiaotong University)
ClassificationAnomaly DetectionGraph Neural NetworkTransformerContrastive LearningGraphStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Propose an energy-based Transformer framework named Controlled Dynamics Attractor Transformer (CDAT), which combines mixed von Mises-Fisher (Mo-vMF) attention energy, Hopfield refinement energy, and CANN-inspired excitation-inhibition modulation to achieve controllable attractor dynamics; achieves state-of-the-art performance on graph classification and graph anomaly detection tasks.
Controlled LLM Training on Spectral Sphere
Tian Xie (Microsoft Research Asia), Baining Guo (Microsoft Research Asia)
OptimizationComputational EfficiencyTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: Designed and implemented the Spectral Sphere Optimizer (SSO), an optimizer that enforces strict weight and update constraints on the Spectral Sphere, to meet the width invariance requirements of Maximal Update Parametrization (μP), and integrated it into Megatron-LM for large-scale LLM pretraining.
Controlled SDEs for Long-Horizon Motion Generation under Latent Decision Uncertainty
Han Zhang (Zhejiang University), Nenggan Zheng (Zhejiang University)
GenerationData SynthesisRobotic IntelligenceTransformerDiffusion modelScore-based ModelFlow-based ModelTime SeriesSequentialBiomedical DataStochastic Differential Equation
🎯 What it does: Studied long-term biological motion prediction under external instruction driving, proposing the CogSDE controlled SDE model
Controlling the Risk of Corrupted Contexts for Language Models via Early-Exiting
Andrea Wynn (Johns Hopkins University), Eric Nalisnick (Johns Hopkins University)
Safty and PrivacyExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose a framework that combines distribution-agnostic risk control with dynamic early exit, enabling the safe use of large language models when harmful or beneficial contexts appear
ConTSG-Bench: A Unified Benchmark for Conditional Time Series Generation
Shaocheng Lan (ShanghaiTech University), Kan Ren (ShanghaiTech University)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringContrastive LearningTextTabularTime SeriesBenchmarkRetrieval-Augmented Generation
🎯 What it does: Built a unified benchmark, ConTSG-Bench, to evaluate the performance of conditional time series generation models under different condition modalities (class labels, attributes, text) and semantic abstraction levels (morphology, concepts).
Convergence Analysis of Decentralized Hessian-/Jacobian-Free Algorithm for Nonconvex Stochastic Bilevel Optimization
Yihan Zhang (Temple University), Hongchang Gao (Temple University)
OptimizationFederated LearningDiffusion modelScore-based ModelAuto EncoderContrastive LearningImageTabularBenchmarkStochastic Differential Equation
🎯 What it does: Developed a decentralized full first-order single-loop algorithm specifically for solving lower-level non-convex but satisfying Polyak-Łojasiewicz (PL) condition bi-level optimization problems.
Convergence Analysis of the Lion Optimizer in Centralized and Distributed Settings
Wei Jiang (Nanjing University of Science and Technology), Lijun Zhang (Nanjing University)
OptimizationFederated LearningComputational EfficiencyContrastive LearningImage
🎯 What it does: This paper theoretically analyzes the convergence properties of the Lion optimizer in both centralized and distributed environments, and proposes a Lion-VR version with variance reduction and a communication-efficient distributed Lion with bidirectional 1-bit sign compression.
Convergence of Steepest Descent and Adam under Non-Uniform Smoothness
Sharan Vaswani (Simon Fraser University), Reza Babanezhad Harikandeh
Optimization
🎯 What it does: This paper theoretically analyzes and proves the convergence of first-order optimization methods such as gradient descent, RMSProp, and Adam under the non-uniform smoothness (H,H0,1)-NS condition, providing linear convergence rates and comparing them with the lower bounds of traditional methods.
Convergence of Two-Timescale Markovian Stochastic Approximations with Applications in Reinforcement Learning
Vagul Mahadevan (Metron), Shangtong Zhang (University of Virginia)
OptimizationReinforcement LearningStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Studied the convergence of stochastic approximation algorithms under two-time-scale Markov noise, proving stability and almost sure convergence, and applied the results to the convergence of TDC(λ);
Convergence Rate Analysis of the AdamW-Style Shampoo: Unifying One-Sided and Two-Sided Preconditioning
Huan Li (Nankai University), Zhouchen Lin (Peking University)
OptimizationLarge Language ModelContrastive LearningTextReview/Survey Paper
🎯 What it does: This paper provides a non-convex convergence rate analysis of the AdamW-style Shampoo optimizer (compatible with both one-sided and two-sided preconditioning).
Convergence Rate of the Last Iterate of Stochastic Proximal Algorithms
Kevin Kurian Thomas Vaidyan (University of British Columbia), Ahmet Alacaoglu (University of British Columbia)
OptimizationFederated LearningContrastive LearningImageTabularBenchmarkStochastic Differential Equation
🎯 What it does: Analyzed two classical algorithms for solving additive composite convex optimization problems, with a focus on relaxing the common bounded variance assumption, proving that under component convexity and smoothness, the final iteration convergence rate of both algorithms is O(1/√T).
Convergent World Representations and Divergent Tasks
Core Francisco Park (Harvard University)
Representation LearningData-Centric LearningTransformerContrastive LearningWorld ModelTabular
🎯 What it does: Studied the impact of multi-task pre-training on the internal world representations of neural networks, and examined the fine-tuning adaptability when new entities are introduced.
Conversation for Non-verifiable Learning: Self-Evolving Large Language Models through Meta-Evaluation
Yuan Sui (National University of Singapore), Bryan Hooi (National University of Singapore)
Meta LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningAgentic AIPrompt EngineeringDiffusion modelScore-based ModelGenerative Adversarial NetworkContrastive LearningTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose a multi-agent self-dialogue framework called CoNL, which enhances both generation and evaluation capabilities through critique and revision among agents, particularly for unverifiable tasks.
Convex Basins in Single-Index Model Loss Landscapes: Applications to Robust Recovery under Strong Adversarial Corruption
SANTANU DAS, jatin batra
OptimizationFederated LearningData-Centric LearningContrastive LearningGaussian SplattingTabularTime SeriesStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Studies the problem of robustly learning Gaussian single-index models (SIMs) under heavy-tailed noise and strong adversarial contamination, and proposes the first robust recovery algorithm tailored for general non-monotonic link functions.
Convex Dataset Valuation for Post-Training
Siqi Zeng (University of Illinois Urbana-Champaign), Xue Feng (Meta)
OptimizationData-Centric LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelContrastive LearningTextMultimodality
🎯 What it does: This paper proposes a convex dataset valuation method based on gradient space, used to select the most valuable auxiliary datasets during the post-training phase of LLMs.
Convex Distance Operator Transport: A Convex and Geometry-Preserving Formulation
Junhyoung Chung (KRAFTON), Gunwoong Park (Seoul National University)
OptimizationRepresentation LearningData-Centric LearningPoint CloudGraphBiomedical Data
🎯 What it does: Propose the Convex Distance Operator Transport (CDOT) framework, which aligns heterogeneous metric-measure spaces (mm space) by aligning distance operators with conditional expectation operators in the L² space, and obtains the optimal transport plan through convex optimization;
Convex Low-resource Accent-Robust Language Detection in Speech Recognition
Miria Feng (Stanford University), Mert Pilanci (Stanford University)
RecognitionOptimizationExplainability and InterpretabilityComputational EfficiencyConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningAudio
🎯 What it does: Propose the Convex Language Detection (CLD) framework, achieving robust language detection in low-resource and accent-diverse speech recognition tasks through convex optimization;
Convex Optimization for Alignment and Preference Learning on a Single GPU
Miria Feng (Stanford University), Mert Pilanci (Stanford University)
OptimizationComputational EfficiencyReinforcement Learning from Human FeedbackTransformerLarge Language ModelAuto EncoderContrastive LearningText
🎯 What it does: This paper proposes COALA, a single GPU language model preference alignment method based on convex optimization.
ConvexBench: Can LLMs Recognize Convex Functions?
Yepeng Liu (University of California Santa Barbara), Yuheng Bu (University of California Santa Barbara)
OptimizationTransformerLarge Language ModelAgentic AIPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Constructed an scalable, mechanically verifiable benchmark ConvexBench to test large language models (LLMs) in identifying convex functions within deep function composition.
Convolutional Learnable-Group Weightless Neural Network
Qinhong Ma (Xiamen University), Bo Mao (Xiamen University)
ClassificationComputational EfficiencyConvolutional Neural NetworkAuto EncoderContrastive LearningImageAudio
🎯 What it does: Propose the CLGN model, a weight-free neural network based on LUT, which achieves end-to-end binary inference in modules such as convolutional layers, threshold pooling, and learnable GroupSum.
Cooperative Variance Estimation and Bayesian Neural Networks for Disentangling Aleatoric and Epistemic Uncertainties
Jiaxiang Yi (Delft University of Technology), Miguel Anibal Bessa (Brown University)
Data-Centric LearningImageTabularBiomedical DataBenchmarkPhysics Related
🎯 What it does: Propose a method that collaborates training of a variance estimation network and a Bayesian neural network to separate the model's aleatoric uncertainty and epistemic uncertainty, thereby improving the mean prediction performance.
CoopEval: Benchmarking Cooperation-Sustaining Mechanisms and LLM Agents in Social Dilemmas
Emanuel Tewolde (Carnegie Mellon University), Zhijing Jin (University of Toronto)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Designed and implemented the CoopEval benchmark suite, conducting large-scale cross-play among six large language models (LLMs) in four classic social dilemmas (prisoner's dilemma, traveler's dilemma, trust game, public goods game) with four types of cooperation mechanisms (repeated games, reputation, mediation, contracts), and evaluated the effectiveness of cooperation through mean, evolutionary fitness, and deviation rating.
Coordinated Disentanglement with Iterative Mode Discovery Under Hidden Correlations
Rong Hu (Zhejiang University), Ling Chen (Zhejiang University)
Domain AdaptationRepresentation LearningGenerative Adversarial NetworkContrastive LearningImageMultimodalityTabularTime Series
🎯 What it does: Propose the CoDID framework, which decomposes hidden-related attributes by utilizing iterative pattern discovery and conditional mutual information minimization.
CooT: Learning to Coordinate In-Context with Coordination Transformers
Huai-Chih Wang (National Taiwan University), Shao-Hua Sun (National Taiwan University)
Meta LearningTransformerSupervised Fine-TuningReinforcement LearningSequentialBenchmark
🎯 What it does: Propose a coordination model called COOT based on Transformer, which achieves instant gradient-free adaptation to unfamiliar partners by utilizing context (past interaction trajectories);
CoPE: A Framework for Optimizing Coordination between Planning and Execution in LLM-based Agents
Huanxi Liu (National University of Defense Technology), Huaimin Wang (National University of Defense Technology)
Autonomous DrivingOptimizationRobotic IntelligenceReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AIMixture of ExpertsTextSequential
🎯 What it does: Propose the CoPE framework, which utilizes self-improving MCTS to collect high-quality plan-execution data, evaluates the executability of planning and the adherence of execution, and uses it as sample weights for LLM fine-tuning.
CoPE: Continual Probe-guided Expansion for Large Vision-Language Models
Ziqin Wang (Beihang University), Si Liu (Beihang University)
Federated LearningComputational EfficiencyKnowledge DistillationRepresentation LearningTransformerMixture of ExpertsVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Proposed a continuous learning framework called CoPE, aimed at addressing the issues of parameter growth and knowledge forgetting in large-scale vision-language models when handling sequential tasks.
COPF: An Online Framework for Deployment-Stable Counterfactual Fairness in Evolving Graphs
Sheng'en Li (Duke Kunshan University), Dongmian Zou (Duke Kunshan University)
Recommendation SystemOptimizationFederated LearningExplainability and InterpretabilityComputational EfficiencyGraph Neural NetworkTransformerReinforcement LearningPrompt EngineeringAuto EncoderGenerative Adversarial NetworkContrastive LearningGraphTime SeriesSequentialBenchmark
🎯 What it does: Proposes COPF, an online framework for achieving deployment-time stable counterfactual fairness on evolving graphs.
Copula-SVI: Vine-Copula Variational Inference with Stein Refining for Instance-Level Correlation Capturing
Junxi Xiao (Sun Yat-sen University), Qinliang Su (Sun Yat-sen University)
Anomaly DetectionOptimizationRepresentation LearningTabularTime Series
🎯 What it does: Designed a Copula-SVI method that decomposes the variational posterior into marginal distributions and regular vine copula, utilizing SVGD to refine particles in the joint latent space, achieving scalable modeling of large-scale instance-level correlations.
Copyright-Bench: Agentic Evaluation of Copyright Law Compliance
Zheng Hui (University of Cambridge), Noam Kolt (Hebrew University)
TransformerLarge Language ModelAgentic AIPrompt EngineeringVision Language ModelImageTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Created the Copyright-Bench benchmark to evaluate whether LLM agents comply with U.S. copyright law in real business workflows;
CORAL: Uncertainty-Aware Regulation of Exposure Concentration in Recommender Systems
Nitin Bisht (University of Technology), Guandong Xu (Education University of Hong Kong)
Recommendation SystemOptimizationExplainability and InterpretabilityTransformerReinforcement LearningMixture of ExpertsContrastive LearningTabularSequential
🎯 What it does: This paper proposes the CORAL framework, which provides interpretable, model-free risk-aware regulation of exposure concentration in recommendation systems by constructing a self-exciting intensity model and an upper confidence bound.
CORE-MTL: Rethinking Gradient Balancing via Causal Orthogonal Representations
Chengfeng Wu (Tsinghua University), Jingge Wang (Tsinghua University)
SegmentationDepth EstimationDomain AdaptationRepresentation LearningConvolutional Neural NetworkAuto EncoderGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: Propose a multi-task learning framework called CORE-MTL based on causal orthogonal representation, focusing on reducing interference between tasks and improving OOD robustness by structurally decomposing the shared representation into semantic flow and residual flow.
CoRe: Collaborative Reasoning via Cross Teaching
Kshitij Mishra (Mohamed bin Zayed University of Artificial Intelligence), Salem Lahlou (Mohamed bin Zayed University of Artificial Intelligence)
Computational EfficiencyKnowledge DistillationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringTextChain-of-Thought
🎯 What it does: Propose a training-time collaborative framework called CORE, which enables multi-model collaborative reasoning during training by allowing failed models to receive hints from successful peers in two rounds of sampling.
CoRe: Combined Rewards with Vision-Language Model Feedback for Preference-Aligned Reinforcement Learning
Hexian Ni (State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences), Yinghao Cai (State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences)
Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningPrompt EngineeringVision Language ModelImageVideoTextChain-of-Thought
🎯 What it does: Proposed the CoRe framework, which decomposes reinforcement learning rewards into formal rewards (FR) and residual rewards (RR), and achieves preference alignment learning without manual annotation through feedback from LLM and VLM.
CORE: Conflict-Oriented Reasoning for General Multimodal Manipulation Detection
Jinjie Shen (Hefei University of Technology), Zhun Zhong (Hefei University of Technology)
Anomaly DetectionExplainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringContrastive LearningImageTextMultimodality
🎯 What it does: Construct a Conflict Attribution Corpus (CAC) and train with conflict awareness to enable multimodal large language models to capture and reason about conflicts, thereby achieving rapid detection of multimodal fake news.
CORE: Context-Robust Remasking for Diffusion Language Models
Kevin Zhai (University of Central Florida), Mubarak Shah (University of Central Florida)
GenerationAI Code AssistantTransformerLarge Language ModelPrompt EngineeringDiffusion modelTextSequentialBenchmark
🎯 What it does: Proposes a training-agnostic context-robust re-masking framework called CORE, which selects and corrects vulnerable tokens by stress-testing the context sensitivity of already generated tokens in masked diffusion models, thereby improving the generation quality.
Correct Looks Better: Pairwise Comparisons Reveal Accuracy Rankings
Mina Remeli (Max Planck Institute for Intelligent Systems), Moritz Hardt (Max Planck Institute for Intelligent Systems)
TransformerLarge Language ModelContrastive LearningTextBenchmark
🎯 What it does: By converting traditional labeled evaluation benchmarks into unlabeled free-form generation evaluations, this study investigates the correlation between these pairwise comparisons generated by LLMs and the actual accuracy rankings using aggregation algorithms such as Bradley-Terry/Elo.
CORRECT: COndensed eRror RECognition via knowledge Transfer in multi-agent systems
Yifan Yu (University of Illinois Urbana Champaign), Bryan Wang (Amazon)
Data SynthesisAnomaly DetectionRepresentation LearningTransformerLarge Language ModelPrompt EngineeringContrastive LearningTextSequentialRetrieval-Augmented Generation
🎯 What it does: Proposed the CORRECT framework, which utilizes compressed error patterns from online caching to achieve error identification in multi-agent systems without training, and released a large-scale CORRECT-Error dataset.
Corrected Samplers for Discrete Flow Models
Zhengyan Wan (East China Normal University), Guang Cheng (University of California Los Angeles)
GenerationData SynthesisDiffusion modelScore-based ModelFlow-based ModelImageTextStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper proposes a time-corrected sampler and a position-corrected sampler for discrete flow models to reduce discretization errors;
Correcting in Hindsight: Editing Past Key-Value States for Robust LLM Reasoning
Mengfei Zhang (Zhejiang University), Leijing Zhou (Zhejiang University)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose HEdit, which corrects early errors in autoregressive inference by real-time detection of trigger points and backtracking to edit key value caches.
Correcting Overparameterization Effects in Fair Empirical Risk Minimization
Xiaoyi MAI, Jean-Michel Loubes (INRIA)
OptimizationFederated LearningExplainability and InterpretabilityRepresentation LearningData-Centric LearningContrastive LearningImageTabular
🎯 What it does: Conduct precise high-dimensional theoretical analysis of fair empirical risk minimization (Fair ERM) for over-parameterized machine learning models, proposing to recover fairness through adjustable bias overcompensation.
Correcting Split Selection in Online Decision Trees via Anytime-Valid Inference
Salim I. Amoukou (J P Morgan AI Research), Manuela Veloso (J P Morgan AI Research)
ClassificationData SynthesisAnomaly DetectionOptimizationFederated LearningComputational EfficiencyAdversarial AttackReinforcement LearningContrastive LearningTabularTime SeriesSequential
🎯 What it does: Propose an online decision tree splitting selection method based on any-time inference, replacing the traditional Hoeffding tree, enabling strict statistical error probability control under any data stream (non-stationary, dependent samples).
Correcting Visual Blur Induced by Attention Distraction to Reduce Hallucinations: Algorithm and Theory
Quanjiang Li (National University of Defense Technology), Chenping Hou (National University of Defense Technology)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Studied the hallucination problem in multi-modal large language models (MLLM) and proposed an untrained Attention-Focused Approach for Improved Image Perception (AFIP) method to correct visual blurriness, enhance visual localization, and significantly reduce hallucination generation.
CorrectionPlanner: Self-Correction Planner with Reinforcement Learning in Autonomous Driving
Yihong Guo (Johns Hopkins University), Xianming Liu (XPENG Motors)
Autonomous DrivingTransformerReinforcement LearningWorld ModelPoint Cloud
🎯 What it does: Proposed a planner called CorrectionPlanner that performs self-correction in the motion token space, capable of proactively identifying and correcting unsafe actions before execution;
Correctness-Optimized Residual Activation Lens (CORAL): Transferrable and Calibration-Aware Inference-Time Steering
Miranda Muqing Miao (University of Pennsylvania), Lyle Ungar (University of Pennsylvania)
Explainability and InterpretabilityComputational EfficiencyKnowledge DistillationRepresentation LearningTransformerLarge Language ModelPrompt EngineeringContrastive LearningTextBenchmark
🎯 What it does: Propose the CORAL method, which trains a regularized MLP probe on internal activations of the LLM to predict residual correctness and perform steering during inference, thereby simultaneously improving accuracy and calibration.
Correspondence Cognitive Learning for Multi-Modal Object Re-Identification
Chao Su (Sichuan University), Yuan Sun (Sichuan University)
RecognitionRetrievalTransformerPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Proposes a Correspondence Cognitive Learning (CCL) framework for multi-modal object re-identification tasks;
Corrigibility Transformation: Constructing Goals That Accept Updates
Rubi Hudson (University of Toronto)
Safty and PrivacyReinforcement LearningTextTabularBenchmark
🎯 What it does: By transforming the reward function, AI objectives are made corrigible (i.e., acceptable for updates) without compromising the original objective's performance.
CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Features
Seonglae Cho (Holistic Ai), Adriano Koshiyama (Holistic Ai)
OptimizationExplainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelPrompt EngineeringAuto EncoderTextBenchmark
🎯 What it does: Propose a model optimization method called CorrSteer, which does not require a contrastive dataset or backpropagation, and instead utilizes features from sparse autoencoders (SAE) activated during generation, and automatically selects features through a two-stage process of correlation plus intervention.
Corruption-Tolerant Asynchronous Q-Learning with Near-Optimal Rates
Sreejeet Maity (North Carolina State University), Aritra Mitra (North Carolina State University)
Reinforcement LearningMixture of ExpertsContrastive LearningTabularSequential
🎯 What it does: Proposes a robust asynchronous Q-learning algorithm that evaluates its performance through convergence rates within finite time and information-theoretic lower bounds under infinite-horizon discounted MDPs where the reward distribution may be heavy-tailed and adversarially contaminated;
Cost-aware Stopping for Bayesian Optimization
Qian Xie (Cornell University), Ziv Scully (Cornell University)
OptimizationHyperparameter SearchNeural Architecture SearchReinforcement LearningContrastive LearningGaussian SplattingTabularTime SeriesSequentialBenchmark
🎯 What it does: Proposes an adaptive stopping rule suitable for cost-aware Bayesian optimization — the PBGI/LogEIPC stopping rule — and verifies its effectiveness in various experiments.
CoT is Not the Chain of Truth: An Empirical Internal Analysis of Reasoning LLMs for Fake News Generation
Zhao Tong (Institute of Information Engineering Chinese Academy Of Sciences), Xiao-Yu Zhang (Institute of Information Engineering Chinese Academy Of Sciences)
Safty and PrivacyExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextChain-of-Thought
🎯 What it does: Studied the security risks of Chain-of-Thought (CoT) internal reasoning in large language models when generating fake news, finding that even if the model refuses the request in the final output, the internal reasoning process may still contain harmful information.
Counterfactual Bootstrap for Robust Meta-Reinforcement Learning
Ai Bo (Syracuse University), M. Cenk Gursoy (Syracuse University)
Meta LearningReinforcement LearningMixture of ExpertsTabular
🎯 What it does: Proposed a counterfactual Bootstrap method for training Meta-RL models on observational data with unmeasured confounding, addressing the reliance of traditional Meta-RL on the no-confounding assumption.
Counterfactual Occlusion-Aware Learning via Visibility Intervention for LiDAR Anomaly Detection
Longyu Yang (University of Electronic Science and Technology of China), Ping Hu (University of Electronic Science and Technology of China)
Anomaly DetectionAutonomous DrivingTransformerDiffusion modelAuto EncoderContrastive LearningPoint Cloud
🎯 What it does: A LiDAR anomaly detection framework based on contrastive causal intervention, named COVAL, is constructed. It utilizes physically aligned synthetic anomalies and controllable occlusions to generate causal contrastive samples, thereby explicitly modeling the relationship between occlusion missing and anomaly structures during training.
Counterfactual Residual Data Augmentation for Regression
Hossein Mohebbi (University of Waterloo), Pascal Poupart (University of Waterloo)
Data SynthesisHyperparameter SearchData-Centric LearningConvolutional Neural NetworkRecurrent Neural NetworkTransformerDiffusion modelGenerative Adversarial NetworkContrastive LearningTabular
🎯 What it does: Proposed and implemented a table regression data augmentation method called CRDA based on residual invariance and counterfactual reasoning;
CountsDiff: A diffusion model on the natural numbers for generation and imputation of count-based data
Renzo Soatto, Maria Skoularidou (Massachusetts Institute of Technology)
GenerationData SynthesisTransformerDiffusion modelScore-based ModelImageTabularBiomedical Data
🎯 What it does: Proposes CountsDiff, a diffusion model that operates on the set of natural numbers, for generating count data and imputing biological count data such as single-cell RNA-seq.
Coupled Cluster con MoLe: Molecular Orbital Learning for Neural Wavefunctions
Luca Thiede (University of Toronto), Alan Aspuru-Guzik
Drug DiscoveryGraph Neural NetworkTransformerSupervised Fine-TuningContrastive LearningGraphTabularPhysics Related
🎯 What it does: Designed and trained an equivariant neural network called M¯oLe, which directly predicts single and double excitation amplitudes of the Coupled Cluster (CC) system from molecular orbitals, significantly reducing the computational cost of CC while maintaining high accuracy.
Coupled Training with Privileged Information and Unlabeled Data
Jiahao Shi (Princeton University), Jason Matthew Klusowski
OptimizationKnowledge DistillationData-Centric LearningContrastive LearningTabular
🎯 What it does: Proposes a joint training framework that leverages privileged information available during training and unlabeled samples to simultaneously learn the deployment model and the rich-view model, thereby improving prediction performance without using privileged information.
Coupled Trigger Optimization and Vulnerable Parameter Alignment for Persistent Backdoor Attacks on Federated Learning
Zhixuan Ma (Xidian University), Han Yu (Nanyang Technological University)
Federated LearningAdversarial AttackContrastive LearningImage
🎯 What it does: Proposes a method to achieve persistent backdoor attacks in federated learning by coupling trigger optimization with fragile parameter alignment.
Coupled Variational Reinforcement Learning for Language Model General Reasoning
Xueru Wen (University of Chinese Academy of Sciences), Debing Zhang (Xiaohongshu Inc)
Computational EfficiencyRepresentation LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningPrompt EngineeringMixture of ExpertsText
🎯 What it does: This study proposes the Coupled Variational Reinforcement Learning (CoVRL) method, which combines prior (question-only) and posterior (answer-guided) generation modes, training the reasoning ability of large language models through a mixture distribution and mixed sampling.
Courtroom Analogy: New Perspective on Uncertainty-Aware Classification
Taeseong Yoon (Korea Advanced Institute of Science and Technology), Heeyoung Kim (Korea Advanced Institute of Science and Technology)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkMixture of ExpertsImage
🎯 What it does: Proposed a single-channel uncertainty estimation framework based on analogies from courtroom debates, and implemented the Mixture of Dirichlet Experts (MoDEX) model.
Covariance estimation using Markov chain Monte Carlo
Yunbum Kook (Georgia Institute of Technology), Matthew Shunshi Zhang
OptimizationFederated LearningComputational EfficiencyRepresentation LearningReinforcement LearningMixture of ExpertsContrastive LearningReview/Survey PaperStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper studies the non-asymptotic estimation of the mean and covariance of a target distribution using Markov chain Monte Carlo (MCMC) methods under the conditions of satisfying the Poincaré inequality and spectral gap, and provides corresponding upper bounds on sample complexity and query complexity.
Covariance Volume Maximization for Embodied Latent Exploration in Deep Reinforcement Learning
Yiming Wang (University of Macau), Leong Hou U (University of Macau)
OptimizationReinforcement LearningContrastive LearningPoint CloudTabularTime SeriesSequential
🎯 What it does: Proposes the Covariance Volume Maximization (CVM) framework, which combines a policy-mixing based behavioral encoder with an exploration reward based on covariance volume expansion, for potential space exploration in deep reinforcement learning.
Coverage ≠ Exposure: Auditable Control of Same-Support Tail Failures under Multimodal Missingness
Ziteng Hong (Beijing Institute of Technology), Guangming Lu (Harbin Institute of Technology)
OptimizationExplainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodalityAudio
🎯 What it does: Proposes a co-support tail failure analysis for multi-modal models under conditions of observable missing and degraded data, and designs an auditable TailPressure metric and an H-CES feedback controller to regulate the exposure of parameter groups during training, thereby enhancing tail robustness.
Coverage Improvement and Fast Convergence of On-policy Preference Learning
Juno Kim (UC Berkeley), Kwang-Sung Jun (POSTECH)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningContrastive LearningText
🎯 What it does: Proposed and proved the 'Coverage Improvement Principle', explaining why using online direct preference optimization (DPO) in policy alignment can significantly improve sample efficiency and achieve rapid convergence;
Coverage, Not Averages: Semantic Stratification for Trustworthy Retrieval Evaluation
Andrew Klearman (Scale AI), Yuan Xue (Scale AI)
RetrievalTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: This paper studies the issue of query distribution imbalance in retrieval evaluation leading to distortion of average metrics, and proposes a semantics-based hierarchical evaluation framework that systematically generates missing queries to ensure coverage.
CoverPruneGS: Coverage-Preserving Structured Pruning for Hierarchical 3D Gaussian Splatting from Sparse-View Monocular Videos
Yang Xiao (University of Technology Sydney), Wenjing Jia (University of Technology Sydney)
CompressionComputational EfficiencyAuto EncoderContrastive LearningGaussian SplattingOptical FlowVideoPoint Cloud
🎯 What it does: A two-level structured pruning framework called CoverPruneGS was developed, specifically for sparse-view monocular video training of hierarchical 3D Gaussian Splatting (3DGS) models, significantly compressing the number of Gaussians while maintaining geometric coverage.
CPMöbius: Iterative Coach–Player Reasoning for Data-Free Reinforcement Learning
Ran Li (Tsinghua University), Maosong Sun (Tsinghua University)
OptimizationData-Centric LearningTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Propose a collaborative Coach-Player framework (CPMobius) to achieve data-free reinforcement learning for enhancing the mathematical reasoning ability of large language models.
CRAG: Can 3D Generative Models Help 3D Assembly?
Zeyu Jiang (New York University), Jing Zhang (New York University)
GenerationData SynthesisTransformerDiffusion modelFlow-based ModelAuto EncoderGenerative Adversarial NetworkContrastive LearningImagePoint CloudMesh
🎯 What it does: Propose a unified framework CRAG that jointly performs 3D fragment assembly and complete shape generation.
Cram Less to Fit More: Training Data Pruning Improves Memorization of Facts
Jiayuan Ye (Apple), Kunal Talwar (Apple)
Computational EfficiencyKnowledge DistillationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringContrastive LearningTextRetrieval-Augmented Generation
🎯 What it does: Improve the language model's factual memory capability through training data screening.
CRAMER: Control via Request-Aware Masking for Editing Recommenders
Zhiyuan Su (Renmin University of China), Ga Wu (Dalhousie University)
Recommendation SystemTransformerPrompt EngineeringContrastive LearningTextSequential
🎯 What it does: Propose the CRAMER framework, which utilizes natural language requests to instantly adjust the parameters of frozen sequential recommendation models, achieving re-adjustment without training through sparse row-column masks;
Creat3r: Confidence Reaggregation for Exploration-aware Active 3D Reconstruction
Chih-Jung Tsai (National Tsing Hua University), Tyng-Luh Liu (Academia Sinica)
OptimizationComputational EfficiencyNeural Radiance FieldGaussian SplattingSimultaneous Localization and MappingOptical FlowImagePoint Cloud
🎯 What it does: Propose Creat3r, an iterative next best view selection framework based on 3D Gaussian Splatting, which constructs a lightweight geometric proxy and guides view selection through confidence and exploration maps.
Credibility-Aware Weighting Federated Causal Discovery for Time Series
Jiegang Xu (Shanxi University), Jiye Liang (Shanxi University)
Federated LearningSafty and PrivacyExplainability and InterpretabilityDiffusion modelScore-based ModelContrastive LearningGaussian SplattingTabularTime SeriesBiomedical Data
🎯 What it does: Propose the Fed-CAW framework to achieve privacy-preserving temporal federated causal discovery using credibility-weighted aggregation.
Credible Information Subset Decomposition: An End-to-End Multi-fidelity Learning Model by Modeling Label Information
Sihan Wang (University of Science and Technology of China), Yang Wang (University of Science and Technology of China)
OptimizationRepresentation LearningDrug DiscoveryGraph Neural NetworkAuto EncoderContrastive LearningGraphTabularPhysics Related
🎯 What it does: Proposed a trustworthy subset decomposition framework that effectively integrates multi-fidelity data through an end-to-end model;
Credit Assignment via Neural Manifold Noise Correlation
Byungwoo Kang (Harvard Medical School), Bernardo L. Sabatini (Harvard University)
ClassificationOptimizationComputational EfficiencyRepresentation LearningConvolutional Neural NetworkRecurrent Neural NetworkAuto EncoderContrastive LearningImageSequential
🎯 What it does: Achieve scalable credit assignment by injecting noise onto a low-dimensional neural manifold and performing noise correlation estimation;
Credit-assigned Policy Gradient for Early Stage Retrieval in Two-stage Ranking
Haruka Kiyohara (Cornell University), Udi Weinsberg (Meta)
Recommendation SystemReinforcement LearningMixture of ExpertsVideoTabular
🎯 What it does: Studied the policy gradient training of early retrievers in two-stage retrieval, proposing a credit assignment policy gradient (CA-PG) method to reduce variance and address the credit assignment problem.
CREDIT: Certified Ownership Verification of Deep Neural Networks Against Model Extraction Attacks
Bolin Shen (Florida State University), Yushun Dong (Florida State University)
Safty and PrivacyAdversarial AttackConvolutional Neural NetworkGraph Neural NetworkContrastive LearningGaussian SplattingImageGraph
🎯 What it does: Proposes CREDIT, a provable ownership verification framework designed for model extraction attacks;
Crisp: A Spectral-Based Interaction Strategy for Multivariate Time Series Forecasting
Binwu Wang (University of Science and Technology of China), Yang Wang (University of Science and Technology of China)
Anomaly DetectionComputational EfficiencyRepresentation LearningRecurrent Neural NetworkTransformerAuto EncoderContrastive LearningMultimodalityTabularTime SeriesBenchmark
🎯 What it does: Proposed a multivariate time series prediction framework called Crisp based on spectral resonance, which enables information exchange only between variables with consistent frequencies by utilizing spectral priors.
CrispEdit: Low-Curvature Projections for Scalable Non-Destructive LLM Editing
Zarif Ikram (University of Southern California), Paria Rashidinejad (University of Southern California)
OptimizationSafty and PrivacyComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningDiffusion modelContrastive LearningText
🎯 What it does: Propose an scalable LLM editing method called CRISPEDIT, focusing on performing factual or safety edits without compromising the model's original capabilities.
Criterion-Conditional In-Context Learning: Evaluating Criterion-Shift Adaptation in Vision-Language Models
Kaiyun Yang (University of Science and Technology of China), Wei Ge (Megvii Technology Inc)
Domain AdaptationAnomaly DetectionTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelContrastive LearningImageVideoTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose Criterion-Conditional In-Context Learning (CCICL), enabling the model to dynamically adjust the discriminative threshold based on the context while keeping the task semantics fixed, addressing the limitations of the 'task = decision criterion' assumption in traditional ICL.
CriticalKV: Optimizing KV Cache Eviction from an Output Perturbation Perspective
Yuan Feng (University of Science and Technology of China), Xike Xie (University of Science and Technology of China)
CompressionOptimizationComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: To address the compression problem of KV cache in large language models, this paper proposes a key cache entry selection algorithm based on worst-case perturbation constraints from the perspective of output perturbation.
Critique-Guided Distillation for Robust Reasoning via Refinement
Berkcan Kapusuzoglu (Capital One), Sambit Sahu (Capital One)
Explainability and InterpretabilityComputational EfficiencyKnowledge DistillationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Propose a training framework called CGD, which trains large models to directly output improved answers from prompts during inference without generating critiques, thereby enhancing the quality of reasoning.
Cross-Chirality Generalization by Axial Vectors for Hetero-Chiral Protein-Peptide Interaction Design
Ziyi Yang (Tsinghua University), Yanyan Lan (Tsinghua University)
Drug DiscoveryProtein Structure PredictionTransformerDiffusion modelBiomedical Data
🎯 What it does: Leverage axial features injected into E(3)-equivariant networks to construct PepMirror, achieving the generation of heterochiral D-peptide complexes from L-L training data.
Cross-Embodiment Robot Foundation World Models with Latent Actions
Huang Huang (Stanford University), Franziska Meier (Meta FAIR Robotics)
Robotic IntelligenceTransformerSupervised Fine-TuningReinforcement LearningVision-Language-Action ModelAuto EncoderContrastive LearningWorld ModelImageVideo
🎯 What it does: Propose a robot world model (LAC-WM) trained in a unified latent action space, which can be pre-trained on multiple robot entities and quickly transferred to unseen robots.