Skip to yearly menu bar Skip to main content


Show Detail Timezone:
America/Los_Angeles
 
Filter Rooms:  

SUN 21 JUL

MON 22 JUL
midnight
(duration 6.0 hr)
Workshop:
(ends 7:00 AM)
Break:
(ends 12:30 AM)
2:30 a.m.
Break:
(ends 4:00 AM)
6 a.m.
Break:
(ends 6:30 AM)
8:30 a.m.
Reception:
(ends 9:45 AM)
11 p.m.
(ends 9:00 AM)
11:45 p.m.
Remarks:
(ends 12:00 AM)

TUE 23 JUL
1 a.m.
Break:
(ends 1:30 AM)
1:30 a.m.
Orals 1:30-2:30
[1:30] Debating with More Persuasive LLMs Leads to More Truthful Answers
[1:45] Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision
[2:00] A Mechanistic Understanding of Alignment Algorithms: A Case Study on DPO and Toxicity
[2:15] Is DPO Superior to PPO for LLM Alignment? A Comprehensive Study
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] Position: Embracing Negative Results in Machine Learning
[1:45] Position: A Safe Harbor for AI Evaluation and Red Teaming
[2:00] Position: Considerations for Differentially Private Learning with Large-Scale Public Pretraining
[2:15] Position: Beyond Personhood: Agency, Accountability, and the Limits of Anthropomorphic Ethical Analysis
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] LSEnet: Lorentz Structural Entropy Neural Network for Deep Graph Clustering
[1:45] Image Clustering with External Guidance
[2:00] Making Old Things New: A Unified Algorithm for Differentially Private Clustering
[2:15] Pruned Pivot: Correlation Clustering Algorithm for Dynamic, Parallel, and Local Computation Models
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] Genie: Generative Interactive Environments
[1:45] Video-LaVIT: Unified Video-Language Pre-training with Decoupled Visual-Motional Tokenization
[2:00] Video-of-Thought: Step-by-Step Video Reasoning from Perception to Cognition
[2:15] VideoPoet: A Large Language Model for Zero-Shot Video Generation
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] SparseTSF: Modeling Long-term Time Series Forecasting with *1k* Parameters
[1:45] Arrows of Time for Large Language Models
[2:00] Unified Training of Universal Time Series Forecasting Transformers
[2:15] SAMformer: Unlocking the Potential of Transformers in Time Series Forecasting with Sharpness-Aware Minimization and Channel-Wise Attention
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] MorphGrower: A Synchronized Layer-by-layer Growing Approach for Plausible Neuronal Morphology Generation
[1:45] EquiPocket: an E(3)-Equivariant Geometric Graph Neural Network for Ligand Binding Site Prediction
[2:00] Expressivity and Generalization: Fragment-Biases for Molecular GNNs
[2:15] Preference Optimization for Molecule Synthesis with Conditional Residual Energy-based Models
(ends 2:30 AM)
2:30 a.m.
(ends 4:00 AM)
3:30 a.m.
Break:
(ends 5:00 AM)
4:30 a.m.
(ends 6:00 AM)
6 a.m.
7 a.m.
Break:
(ends 7:30 AM)
7:30 a.m.
Orals 7:30-8:30
[7:30] Position: The Platonic Representation Hypothesis
[7:45] Robustness of Nonlinear Representation Learning
[8:00] Provable Multi-Task Representation Learning by Two-Layer ReLU Neural Networks
[8:15] Rejuvenating image-GPT as Strong Visual Representation Learners
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] Position: Technical Research and Talent is Needed for Effective AI Governance
[7:45] Position: AI-Powered Autonomous Weapons Risk Geopolitical Instability and Threaten AI Research
[8:00] Position: Near to Mid-term Risks and Opportunities of Open-Source Generative AI
[8:15] Position: On the Societal Impact of Open Foundation Models
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] How Private are DP-SGD Implementations?
[7:45] Private Truly-Everlasting Robust-Prediction
[8:00] ViP: A Differentially Private Foundation Model for Computer Vision
[8:15] PrE-Text: Training Language Models on Private Federated Data in the Age of LLMs
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] Symbolic Music Generation with Non-Differentiable Rule Guided Diffusion
[7:45] DITTO: Diffusion Inference-Time T-Optimization for Music Generation
[8:00] Fast Timing-Conditioned Latent Audio Diffusion
[8:15] Listenable Maps for Audio Classifiers
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] Transformers Learn Nonlinear Features In Context: Nonconvex Mean-field Dynamics on the Attention Landscape
[7:45] I/O Complexity of Attention, or How Optimal is FlashAttention?
[8:00] Improving Transformers with Dynamically Composable Multi-Head Attention
[8:15] Less is More: on the Over-Globalizing Problem in Graph Transformers
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] Compressible Dynamics in Deep Overparameterized Low-Rank Learning & Adaptation
[7:45] APT: Adaptive Pruning and Tuning Pretrained Language Models for Efficient Training and Inference
[8:00] DiJiang: Efficient Large Language Models through Compact Kernelization
[8:15] Any-Precision LLM: Low-Cost Deployment of Multiple, Different-Sized LLMs
(ends 8:30 AM)
8:30 a.m.
11:30 p.m.
(ends 9:00 AM)

WED 24 JUL
1 a.m.
Break:
(ends 1:30 AM)
1:30 a.m.
Orals 1:30-2:30
[1:30] Position: Automatic Environment Shaping is the Next Frontier in RL
[1:45] Pausing Policy Learning in Non-stationary Reinforcement Learning
[2:00] OMPO: A Unified Framework for RL under Policy and Dynamics Shifts
[2:15] Online Matching with Stochastic Rewards: Provable Better Bound via Adversarial Reinforcement Learning
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution
[1:45] Mean-field Chaos Diffusion Models
[2:00] NaturalSpeech 3: Zero-Shot Speech Synthesis with Factorized Codec and Diffusion Models
[2:15] Scaling Rectified Flow Transformers for High-Resolution Image Synthesis
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] Chain of Code: Reasoning with a Language Model-Augmented Code Emulator
[1:45] SceneCraft: An LLM Agent for Synthesizing 3D Scenes as Blender Code
[2:00] Interpreting and Improving Large Language Models in Arithmetic Calculation
[2:15] Evaluation of LLMs on Syntax-Aware Code Fill-in-the-Middle Tasks
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] Active Statistical Inference
[1:45] Sparse Inducing Points in Deep Gaussian Processes: Enhancing Modeling with Denoising Diffusion Variational Inference
[2:00] Probabilistic Generating Circuits - Demystified
[2:15] Probabilistic Inference in Language Models via Twisted Sequential Monte Carlo
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] Position: Measure Dataset Diversity, Don't Just Claim It
[1:45] Rethinking Data Shapley for Data Selection Tasks: Misleads and Merits
[2:00] Monitoring AI-Modified Content at Scale: A Case Study on the Impact of ChatGPT on AI Conference Peer Reviews
[2:15] Differentiable Mapper for Topological Optimization of Data Representation
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] Doubly Robust Causal Effect Estimation under Networked Interference via Targeted Learning
[1:45] Hybrid$^2$ Neural ODE Causal Modeling and an Application to Glycemic Response
[2:00] Inferring the Long-Term Causal Effects of Long-Term Treatments from Short-Term Experiments
[2:15] ACE: Off-Policy Actor-Critic with Causality-Aware Entropy Regularization
(ends 2:30 AM)
2:30 a.m.
(ends 4:00 AM)
3:30 a.m.
Break:
(ends 5:00 AM)
4:30 a.m.
(ends 6:00 AM)
6 a.m.
7 a.m.
Break:
(ends 7:30 AM)
Town Hall:
(ends 7:30 AM)
7:30 a.m.
Orals 7:30-8:30
[7:30] Offline Actor-Critic Reinforcement Learning Scales to Large Models
[7:45] Stop Regressing: Training Value Functions via Classification for Scalable Deep RL
[8:00] SAPG: Split and Aggregate Policy Gradients
[8:15] Rate-Optimal Policy Optimization for Linear Markov Decision Processes
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] InfoNet: Neural Estimation of Mutual Information without Test-Time Optimization
[7:45] Optimal Hessian/Jacobian-Free Nonconvex-PL Bilevel Optimization
[8:00] Principled Preferential Bayesian Optimization
[8:15] Zeroth-Order Methods for Constrained Nonconvex Nonsmooth Stochastic Optimization
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] Stealing part of a production language model
[7:45] Trained Random Forests Completely Reveal your Dataset
[8:00] AI Control: Improving Safety Despite Intentional Subversion
[8:15] Low-Cost High-Power Membership Inference Attacks
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] Automated Evaluation of Retrieval-Augmented Language Models with Task-Specific Exam Generation
[7:45] MagicLens: Self-Supervised Image Retrieval with Open-Ended Instructions
[8:00] Repoformer: Selective Retrieval for Repository-Level Code Completion
[8:15] Bottleneck-Minimal Indexing for Generative Document Retrieval
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] Position: Do pretrained Transformers Learn In-Context by Gradient Descent?
[7:45] ExCP: Extreme LLM Checkpoint Compression via Weight-Momentum Joint Shrinking
[8:00] How do Large Language Models Navigate Conflicts between Honesty and Helpfulness?
[8:15] Flextron: Many-in-One Flexible Large Language Model
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] Does Label Smoothing Help Deep Partial Label Learning?
[7:45] SAM as the Guide: Mastering Pseudo-Label Refinement in Semi-Supervised Referring Expression Segmentation
[8:00] Candidate Pseudolabel Learning: Enhancing Vision-Language Models by Prompt Tuning with Unlabeled Data
[8:15] Speech Self-Supervised Learning Using Diffusion Model Synthetic Data
(ends 8:30 AM)
11:30 p.m.
(ends 9:00 AM)

THU 25 JUL
midnight
Invited Talk:
Chelsea Finn
(ends 1:00 AM)
1 a.m.
Break:
(ends 1:30 AM)
1:30 a.m.
Orals 1:30-2:30
[1:30] Emergent Equivariance in Deep Ensembles
[1:45] From Coarse to Fine: Enable Comprehensive Graph Self-supervised Learning with Multi-granular Semantic Ensemble
[2:00] Decomposing Uncertainty for Large Language Models through Input Clarification Ensembling
[2:15] AlphaFold Meets Flow Matching for Generating Protein Ensembles
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] On the Last-Iterate Convergence of Shuffling Gradient Methods
[1:45] Multiplicative Weights Update, Area Convexity and Random Coordinate Descent for Densest Subgraph Problems
[2:00] High-Probability Convergence for Composite and Distributed Stochastic Minimization and Variational Inequalities with Heavy-Tailed Noise
[2:15] Information Complexity of Stochastic Convex Optimization: Applications to Generalization, Memorization, and Tracing
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] Position: Rethinking Post-Hoc Search-Based Neural Approaches for Solving Large-Scale Traveling Salesman Problems
[1:45] Evolution of Heuristics: Towards Efficient Automatic Algorithm Design Using Large Language Model
[2:00] S$\Omega$I: Score-based O-INFORMATION Estimation
[2:15] A Dynamic Algorithm for Weighted Submodular Cover Problem
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] PRISE: LLM-Style Sequence Compression for Learning Temporal Action Abstractions in Control
[1:45] Fast Co-Training under Weak Dependence via Stream-Based Active Learning
[2:00] Self-Composing Policies for Scalable Continual Reinforcement Learning
[2:15] Stereo Risk: A Continuous Modeling Approach to Stereo Matching
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] Theoretical Analysis of Learned Database Operations under Distribution Shift through Distribution Learnability
[1:45] Discovering Environments with XRM
[2:00] LCA-on-the-Line: Benchmarking Out of Distribution Generalization with Class Taxonomies
[2:15] Test-Time Model Adaptation with Only Forward Passes
(ends 2:30 AM)
Orals 1:30-2:30
[1:30] Position: Opportunities Exist for Machine Learning in Magnetic Fusion Energy
[1:45] Locality-Sensitive Hashing-Based Efficient Point Transformer with Applications in High-Energy Physics
[2:00] Parameterized Physics-informed Neural Networks for Parameterized PDEs
[2:15] Challenges in Training PINNs: A Loss Landscape Perspective
(ends 2:30 AM)
2:30 a.m.
(ends 4:00 AM)
3 a.m.
(ends 9:00 AM)
3:30 a.m.
Break:
(ends 5:00 AM)
4:30 a.m.
(ends 6:00 AM)
6 a.m.
View of AI from the European Commission:
Lucilla Sioli
(ends 7:00 AM)
7 a.m.
Break:
(ends 7:30 AM)
7:30 a.m.
Orals 7:30-8:30
[7:30] Position: Open-Endedness is Essential for Artificial Superhuman Intelligence
[7:45] Learning to Model the World With Language
[8:00] CompeteAI: Understanding the Competition Dynamics of Large Language Model-based Agents
[8:15] GPTSwarm: Language Agents as Optimizable Graphs
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] Accurate LoRA-Finetuning Quantization of LLMs via Information Retention
[7:45] DoRA: Weight-Decomposed Low-Rank Adaptation
[8:00] GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection
[8:15] LoRA Training in the NTK Regime has No Spurious Local Minima
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] NExT-GPT: Any-to-Any Multimodal LLM
[7:45] MLLM-as-a-Judge: Assessing Multimodal LLM-as-a-Judge with Vision-Language Benchmark
[8:00] FedMBridge: Bridgeable Multimodal Federated Learning
[8:15] A Touch, Vision, and Language Dataset for Multimodal Alignment
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] Learning Useful Representations of Recurrent Neural Network Weight Matrices
[7:45] Data-free Neural Representation Compression with Riemannian Neural Dynamics
[8:00] Neural Collapse meets Differential Privacy: Curious behaviors of NoisyGD with Near-Perfect Representation Learning
[8:15] Contrasting Multiple Representations with the Multi-Marginal Matching Gap
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] The Stronger the Diffusion Model, the Easier the Backdoor: Data Poisoning to Induce Copyright BreachesWithout Adjusting Finetuning Pipeline
[7:45] Robust CLIP: Unsupervised Adversarial Fine-Tuning of Vision Embeddings for Robust Large Vision-Language Models
[8:00] Towards Optimal Adversarial Robust Q-learning with Bellman Infinity-error
[8:15] Scalable AI Safety via Doubly-Efficient Debate
(ends 8:30 AM)
Orals 7:30-8:30
[7:30] Active Adaptive Experimental Design for Treatment Effect Estimation with Covariate Choice
[7:45] All-in-one simulation-based inference
[8:00] Privacy Preserving Adaptive Experiment Design
[8:15] Environment Design for Inverse Reinforcement Learning
(ends 8:30 AM)
8:30 a.m.
11 p.m.
(ends 7:00 AM)
11:30 p.m.
Break:
(ends 12:00 AM)

SAT 27 JUL
3:30 a.m.
Break:
(ends 5:00 AM)
6:30 a.m.
Break:
(ends 7:00 AM)