AIRI на ICLR 2026 🔥

23 апреля в Рио-де-Жанейро, Бразилия, стартовала конференция ICLR — одно из важнейших мероприятий в мире машинного обучения. В этом году исследователи AIRI представляют на ней 31 статью:

⚫️Tighter Performance Theory of FedExProx
⚫️Unraveling the Complexity of Memory in RL Agents: an Approach for Classification and Evaluation
⚫️Obfuscated Activations Bypass LLM Latent-Space Defenses
⚫️A Statistical Learning Perspective on Semi-dual Adversarial Neural Optimal Transport Solvers
⚫️InfoBridge: Mutual Information estimation via Bridge Matching
⚫️Birch SGD: A Tree Graph Framework for Local and Asynchronous SGD Methods
⚫️Locally Subspace-Informed Neural Operators for Efficient Multiscale PDE Solving
⚫️Curse of Slicing: Why Sliced Mutual Information is a Deceptive Measure of Statistical Dependence
⚫️Interaction Field Matching: Overcoming Limitations of Electrostatic Models
⚫️Proving the Limited Scalability of Centralized Distributed Optimization via a New Lower Bound Construction
⚫️LoRA meets Riemannion: Muon Optimizer for Parametrization-independent Low-Rank Adapters
⚫️Speech-to-LaTeX: New Models and Datasets for Converting Spoken Equations and Sentences
⚫️Asynchronous Policy Gradient Aggregation for Efficient Distributed Reinforcement Learning
⚫️Deep Learning for Subspace Regression
⚫️Universal Inverse Distillation for Matching Models with Real-Data Supervision (No GANs)
⚫️Entering the Era of Discrete Diffusion Models: A Benchmark for Schrödinger Bridges and Entropic Optimal Transport
⚫️GAS: Improving Discretization of Diffusion ODEs via Generalized Adversarial Solver
⚫️Q-RAG: Long Context Multi-step Retrieval via Value-based Embedder Training
⚫️MMReD: a Cross-Modal Benchmark for Dense Context Reasoning
⚫️Logit‑KL Flow Matching: Non‑Autoregressive Text Generation via Sampling‑Hybrid Inference
⚫️VINTIX II: Decision Pre-Trained Transformer is a Scalable in-Context Reinforcement Learner
⚫️Modeling the Density of Pixel-level Self-supervised Embeddings for Unsupervised Pathology Segmentation in Medical CT
⚫️ELMUR: External Layer Memory with Update/Rewrite for Long-Horizon RL Problems
⚫️Memory, Benchmark & Robots: A Benchmark for Solving Complex Tasks with Reinforcement Learning
⚫️Recurrent Action Transformer with Memory
⚫️Learning of Population Dynamics: Inverse Optimization Meets JKO Scheme
⚫️GeomMotif: A Benchmark for Arbitrary Geometric Preservation in Protein Generation
⚫️Zero-Shot Adaptation of Behavioral Foundation Models to Unseen Dynamics
⚫️cadrille: Multi-modal CAD Reconstruction with Online Reinforcement Learning
⚫️HOTA: Hamiltonian framework for Optimal Transport Advection
⚫️Diffusion & Adversarial Schrödinger Bridges via Iterative Proportional Markovian Fitting

Делимся фотографиями с первых дней конференции 📷

P. S. Про воркшопы расскажем отдельно.

#AIRIнаКонфе