论文库

Program evolution provides a promising approach to discover agent memory designs and study the choices underlying their effectiveness. We introduce MAPE, a framework that evolves whole memory programs from minimal seeds through a standardized lifecycle interface and an iterative diagnosis-proposal loop. Experiments on LoCoMo and LongMemEval yield an absolute accuracy gain of up to 12% over manually designed baselines at 35% of the serving cost, with further evaluations supporting transfer across serving backbones and benchmarks. Empirical analyses and ablations identify and validate two recurring design patterns: retaining raw conversations as retrievable evidence preserves details omitted from extracted memories, while composing retrieved evidence according to query intent helps the answering model use it effectively. Transplanting the composition module into five existing memory systems improves accuracy across all five. Beyond these patterns, the search objective shapes where computation goes: optimizing serving cost alongside accuracy shifts computation from LLM-based memory construction to selective processing at query time while retaining competitive accuracy. Together, these findings offer empirical lessons for agent memory design. Code is available at https://github.com/ash-ding/MAPE.

Most AI assistants respond only when prompted, remaining silent even when events in the video stream offer meaningful opportunities for conversation. To move beyond purely user-driven interaction, we present AURA 2.0, a streaming video large language model that introduces three core upgrades to AURA. The key advance is Model-Initiated Conversation (MIC), which enables the model to identify topics worth discussing in the evolving visual stream and initiate conversations at appropriate moments without a user query. Stream-Aware Tool Calling (SATC) enables the model to invoke external tools using the current visual and conversational context and incorporate returned information into the ongoing interaction. Interactive Visual Grounding (IVG) supports visual grounding in both user inputs and model outputs. The model can identify objects marked by users on video frames and produce frame-aligned bounding boxes to link its responses to specific objects and regions. To develop these capabilities, we construct a dedicated streaming interaction dataset and adopt a three-stage training scheme to integrate the new behaviors with AURA's existing streaming interaction abilities. We further incorporate Long-Horizon Interaction Memory (LHIM) to maintain contextual continuity over extended video streams. Evaluations on public streaming benchmarks show that AURA 2.0 achieves leading performance among open-source streaming models, while targeted evaluations further validate its new interaction capabilities. We open-source the model weights and inference code.

Developing effective machine learning models requires complex manual tuning, while current Automated Machine Learning (AutoML) frameworks remain domain-specific and isolate model architecture, training hyperparameter, and data ratio optimization. Here, we bridge this gap by proposing a universal, large language model (LLM)-based AutoML framework that jointly optimizes these configurations for the first time without the need for task-specific fine-tuning. We propose NCode, a unified numerical encoding scheme that translates diverse AutoML design choices into universal discrete sequences and effectively reframes the optimization process as a unified ranking task. Trained solely on optimization trajectories derived from a limited, single-domain vision task, the proposed LLM-based searcher demonstrates impressive cross-task, cross-modality and joint search-space optimization generalizability, achieving state-of-the-art performance across image classification, image segmentation, speaker verification and text-to-image benchmarks. Crucially, it exhibits superior scalability in the large-scale training and joint optimization of a billion-parameter Diffusion Transformer, requiring 896 GPU-days and attaining a state-of-the-art GenEval score of 0.91. By leveraging reinforcement learning to instill Chain-of-Thought reasoning, we provide a systematic way to interpret the LLM-based searcher's internal decision-making mechanisms, revealing its optimization strategy shifted from the early-stage diversity-driven exploration to the late-stage anchor exploitation. Our proposed framework provides a scalable, generalizable, and interpretable pathway toward general-purpose open-world machine learning optimization.

AI agents are playing an increasingly important role in physics research. However, existing physics benchmarks only evaluate fixed questions, isolated sub-tasks, or paper reproduction with given solutions, and rarely check research quality alongside correctness. We introduce FrontierPhysics, an end-to-end benchmark with 56 tasks made by 42 physicists based on their own research, covering nine domains across theoretical, computational, and experimental physics. Each task requires agents to review the literature, develop a research plan, conduct research and experiments, and produce structured outputs and scientific manuscripts. Combining deterministic verifiers with expert-authored rubrics, FrontierPhysics evaluates whether agents' research is correct and publishable. Across 22 models in 26 configurations, no configuration passes more than 46 of its 168 rollouts. Manual review of agent trajectories shows that frontier agents' main weaknesses are insufficiently rigorous reasoning and limited expert intuition, rather than gaps in physics knowledge or execution errors.