theoretical guarantees
Formal statements that provide assurances about the performance or behavior of algorithms under certain conditions. These guarantees are essential for understanding the reliability and robustness of AI models.
- $Q\sharp$: Provably Optimal Distributional RL for LLM Post-Training
- A Unified Framework for Provably Efficient Algorithms to Estimate Shapley Values
- A learnability analysis on neuro-symbolic learning
- ADG: Ambient Diffusion-Guided Dataset Recovery for Corruption-Robust Offline Reinforcement Learning
- AdaReasoner: Adaptive Reasoning Enables More Flexible Thinking
- Adaptive and Multi-scale Affinity Alignment for Hierarchical Contrastive Learning
- Angular Constraint Embedding via SpherePair Loss for Constrained Clustering
- Backward Conformal Prediction
- Bilevel Network Learning via Hierarchically Structured Sparsity
- Convergence of Clipped SGD on Convex $(L_0,L_1)$-Smooth Functions
- DAPO : Improving Multi-Step Reasoning Abilities of Large Language Models with Direct Advantage-Based Policy Optimization
- Decreasing Entropic Regularization Averaged Gradient for Semi-Discrete Optimal Transport
- Deep Learning with Plausible Deniability
- Deeper with Riemannian Geometry: Overcoming Oversmoothing and Oversquashing for Graph Foundation Models
- Distances for Markov chains from sample streams
- Distributional Autoencoders Know the Score
- Distributionally Robust Performative Optimization
- Efficient Preference-Based Reinforcement Learning: Randomized Exploration meets Experimental Design
- ElliCE: Efficient and Provably Robust Algorithmic Recourse via the Rashomon Sets
- Evaluating LLM-contaminated Crowdsourcing Data Without Ground Truth
- Exponential Convergence Guarantees for Iterative Markovian Fitting
- FACE: Faithful Automatic Concept Extraction
- Fairness-aware Bayes Optimal Functional Classification
- GVPO: Group Variance Policy Optimization for Large Language Model Post-Training
- Instance-Dependent Regret Bounds for Nonstochastic Linear Partial Monitoring
- Johnson-Lindenstrauss Lemma Beyond Euclidean Geometry
- LLM Interpretability with Identifiable Temporal-Instantaneous Representation
- Learning Repetition-Invariant Representations for Polymer Informatics
- Lyapunov-Stable Adaptive Control for Multimodal Concept Drift
- Measure-Theoretic Anti-Causal Representation Learning
- Model Selection for Off-policy Evaluation: New Algorithms and Experimental Protocol
- NUTS: Eddy-Robust Reconstruction of Surface Ocean Nutrients via Two-Scale Modeling
- NeuralSurv: Deep Survival Analysis with Bayesian Uncertainty Quantification
- New Perspectives on the Polyak Stepsize: Surrogate Functions and Negative Results
- Offline Guarded Safe Reinforcement Learning for Medical Treatment Optimization Strategies
- Optimal Estimation of the Best Mean in Multi-Armed Bandits
- Optimal Online Change Detection via Random Fourier Features
- Oracle-Efficient Combinatorial Semi-Bandits
- Performative Risk Control: Calibrating Models for Reliable Deployment under Performativity
- PhySense: Sensor Placement Optimization for Accurate Physics Sensing
- PhySense: Sensor Placement Optimization for Accurate Physics Sensing
- Privacy amplification by random allocation
- SING: SDE Inference via Natural Gradients
- Safely Learning Controlled Stochastic Dynamics
- Scalable Exploration via Ensemble++
- Scalable Policy-Based RL Algorithms for POMDPs
- Selective Omniprediction and Fair Abstention
- Semantic Representation Attack against Aligned Large Language Models
- Some Optimizers are More Equal: Understanding the Role of Optimizers in Group Fairness
- Sparse Optimistic Information Directed Sampling
- Streaming Attention Approximation via Discrepancy Theory
- TANDEM: Bi-Level Data Mixture Optimization with Twin Networks
- Theoretical Guarantees for the Retention of Strict Nash Equilibria by Coevolutionary Algorithms
- Thought Communication in Multiagent Collaboration
- Tight Generalization Bounds for Large-Margin Halfspaces
- Topology-Aware Conformal Prediction for Stream Networks
- Towards a Pairwise Ranking Model with Orderliness and Monotonicity for Label Enhancement
- Transfer Learning on Edge Connecting Probability Estimation Under Graphon Model
- V-CECE: Visual Counterfactual Explanations via Conceptual Edits
- Which Algorithms Have Tight Generalization Bounds?