…Yi Jing , , , , , , Xiaozhi Wang Abstract SAERL uses Sparse Autoencoder-derived signals from model internals to enhance LLM reinforcement learning through diversity control, difficulty-aware curriculum learning, and quality-based data filtering. AI…
Papers arxiv:2605.02913 Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learning Published on Apr 8 Submitted by Rohan Surana on May 6 McAuley-Lab Authors…
…Generated by Qwen/Qwen2.5-Coder-32B-Instruct Reinforcement learning (RL) for visual reasoning needs scalable, verifiable, and controllable training signals. Existing visual RL post-training trains on static curated datasets, with…
Papers arxiv:2605.11711 Debiased Model-based Representations for Sample-efficient Continuous Control Published on May 12 Submitted by Jiafei Lyu on May 13 Tencent Hunyuan Authors: , , , Kai Yang , , , , Abstract DR.Q…
…No dataset linking this paper Cite arxiv.org/abs/2604.27711 in a dataset README.md to link it from this page. No Space linking this paper Cite arxiv.org/abs/2604…
…Sriram Narayanan , , , Abstract PhyCo enhances video diffusion models with physics-based control through a large-scale dataset, physics-supervised fine-tuning, and vision-language model guidance for improved physical consistency. AI-generated…
…datasets, open-weight action tokenizers, architectural redesign for continuous-action prediction, and adaptive reasoning for reduced latency. AI-generated summary Vision-Language-Action (VLA) models aim to provide a single generalist controller…
…Instance segmentation Balanced 1K / 10K / 100K subsets support controlled scaling-behavior studies, alongside a full 2M-scale corpus for large-scale training. Dataset and code are publicly available on Hugging Face and…
…broad range of controllable generation tasks while preserving modularity, composability , and practical extensibility across rapidly evolving diffusion backbones . All resources will be open sourced, including code, models, and datasets. View arXiv page…
…the discovery environment must make the control space tractable and provide cheap, frequent feedback for TTS search. As a concrete instantiation, we formulate width--depth TTS as controller synthesis over pre-collected…