arXiv 论文速递

Pipeline figures in ML papers must be repurposed across many canvases, including paper columns, 16:9 slides, portrait posters, 1:1 social teasers, 9:16 phone previews. Each format …

In this paper, we study how training data creates associations between the tokens at the start of a base model's response and the reasoning behavior that follows. First, we demonst…

Worst-group accuracy (WGA) evaluates a trained predictor but does not characterize how its frozen backbone behaves when a new head is learned. We introduce BiasFlow, a hook-based t…

揭示多模态扩散 Transformer 中文本与视觉表征的交互机制，为理解并改进生成模型的跨模态对齐提供新视角。

Multimodal Diffusion Transformers (MM-DiTs) jointly process visual and textual representations throughout generation. These models repeatedly update the text tokens through multimo…

让 LLM 智能体通过视频记忆编排冻结的 VLA 策略，实现跨回合自我改进，对具身智能体的持续学习与泛化有重要价值。

LLM agents that orchestrate frozen vision-language-action (VLA) policies improve across episodes through text memory, which records what the agent did but not how the task is done.…

Some diffusion posterior samplers construct Gaussian-tilted intermediate distributions along the reverse process. We observe that these targets can be pulled back to clean-space po…

Every recurrence of a looped language model adds cost in training, decoding, prefill, and reinforcement learning (RL). The closer recurrent states get to fixed points, the less the…

Sliders provide an intuitive interface for continuous image editing. In current generative approaches, however, the slider is simply a rescaling of the method's strength parameter,…

针对 LLM 智能体记忆冗余与检索低效问题，提出按需多模态记忆管理机制，提升长程交互中的信息复用效率。

Memory has become integral to the LLM agent ecosystem, supporting information retention and reuse across interactions. However, most existing agent memory systems construct memory …

用保形预测为网页智能体提供可靠自验证信号，缓解 RL 训练依赖弱监督的问题，提升训练与测试时扩展的可靠性。

Open-source web agents are now strong enough to execute realistic browser tasks, but training them with reinforcement learning still depends on weak supervision: binary task succes…

首次系统评测前沿模型的实验研究品味，即选题与判断能力，为评估 AI 科研自主性提供了新基准。

We introduce TasteVal, a benchmark to evaluate the experimental research taste of frontier models. We define research taste as the ability to pick interesting problems to solve, de…

Large longitudinal cohorts often contain wrist accelerometry without optical heart-rate sensing, motivating recovery of cardiac information from motion signals already collected du…