Skip to main content
Search

Tsinghua University

99 related articles

Subscribe to updates
  1. arXiv

    Chemists write only the target, and VibeSyn's agents run a ten-step total synthesis: active human input is 7.3%

    A team led by Zhejiang University built VibeSyn, a hierarchical agent system that turns a chemist's molecular objective into synthesis decisions revised as experimental evidence accumulates, then drives computational tools and automated devices to execute them. In a ten-step catharanthine total synthesis the system recorded 722 actions, of which agent reasoning was 69.7% and tool or device outcomes 18.8%, while active human input and human confirmation were 7.3% and 4.2%; starting from 29.8 mmol of 3-ethylpyridine it produced 160 mg of product. Across three campaigns synthesizing cyclic peptides containing unnatural amino acids, active human input and confirmation were 7.5% and 5.2%.

  2. arXiv

    Reconstruction residuals miss a drifting sensor: CLOSURE adds a second, separately learned channel-consistency score and reaches 88.13 Aff-F1 on SMD

    Multivariate time series anomaly detection has long leaned on reconstruction residuals, but a residual only certifies that a window can be reproduced, not that its channels still explain one another as they do under normal operation. CLOSURE splits recoverability and cross-channel consistency into two prediction tasks trained separately on normal data: a diffusion reconstructor conditioned on a discrete memory of normal temporal bases scores the first, and a relation operator that predicts each channel only from its peers scores the second, with the two combined after calibration on normal data. Across the five public benchmarks SMD, MSL, SMAP, SWaT and PSM, its Aff-F1 exceeds the listed methods, reaching 88.13 on SMD.

  3. arXiv

    Action generation stops stitching chunks: ODEAction evolves actions by ODE and lifts EBench average success rate by 61.8% relative

    Vision-language-action models usually predict a whole future action chunk in one pass, and adjacent chunks lack explicit temporal consistency constraints, so actions can jump at chunk boundaries. ODEAction recasts generation as continuous evolution from the current action: it learns an action transition field and integrates it with an ordinary differential equation to produce subsequent actions, instead of jointly predicting a fixed-length chunk. Average success rate rises from 96.85% to 98.25% on LIBERO and from 15.28% to 24.72% on EBench, a 61.8% relative gain; on five RoboTwin tasks it reaches 84.4% average success against the 82.4% baseline.

  4. arXiv

    Low-bit agents stop dropping task steps: SteerDistill's closed-loop teacher corrections land within one point of BF16 at W4A16

    Low-bit quantization cuts the inference cost of LLM agents but often makes them overlook task requirements or environmental conditions and take wrong steps in tool use. SteerDistill has a full-precision teacher locate the student's first behavioral divergence, issue a structured correction prompt, and refine that prompt using execution feedback until the revision passes local verification. The student then learns these corrections without guidance, and across five agentic benchmarks and three models under W4A16 its average score exceeds the strongest quantized baseline by 1.67–2.18 points and stays within one point of BF16.

  5. arXiv

    Multi-turn RL pushes LLM agents past the base model only with roomy budgets, and the edge fades or reverses when turns tighten

    Multi-turn agentic reinforcement learning is widely used to strengthen the long-horizon behavior of large language model agents, yet trajectory-level success rates hide what it does to individual decisions. Across four interactive tasks, DeepCoder-MT, BFCL, InterCode-SQL and FrozenLake, models trained with GRPO are evaluated along two axes, sampling breadth and interaction depth: with more sampled trajectories the RL model consistently beats its base model, but when the turns allowed per trajectory are tightened that edge shrinks, disappears, and on some tasks reverses. Matched-state comparisons show better next-step actions on only 46.8% of DeepCoder-MT states and 42.8% of FrozenLake states, with the gain concentrated in error recovery and later trajectory stages.

  6. arXiv

    Visual verifiers stop answering only true or false: OmniVerifier-C1 splits prompts into sub-questions and lifts ViVerBench overall to 0.684

    Visual verifiers have long returned only a true-or-false verdict, which cannot separate partially correct outputs from worse ones. OmniVerifier-C1 decomposes a complex prompt or environmental rule into fine-grained sub-questions, judges each one binarily, and turns the satisfied fraction into a continuous score, while a decoupled reinforcement-learning scheme optimizes sub-question judgments and global alignment scoring together. On ViVerBench, the Qwen3-VL-8B version trained with graded supervised fine-tuning plus graded reinforcement learning reaches an overall score of 0.684, above 0.671 for binary reinforcement learning alone.

  7. arXiv

    Tool outputs no longer have to be text: Coca-Cola compresses search results into latent tokens, cutting a web-search agent's token use by more than 70%

    Each time a web-search agent calls a tool, the long returned text is appended verbatim to its context, so after several steps the context balloons and dilutes attention. Coca-Cola instead compresses those observations into a few continuous latent tokens and embeds them directly in the agent's input embedding space, with action-preserving training aligning the next-action distributions under full and compressed observations. Across four QA benchmarks and two model families, total token consumption falls by 70.01%–89.24% and end-to-end runtime by 24.45%–55.97%, while EM and F1 exceed the uncompressed agent in every setting.

  8. arXiv

    Enzyme–reaction pairing moves beyond sequence similarity: RIVER pools evidence with multiple agents and places all four target enzymes near the top of 667 candidates

    RIVER is a training-free multi-agent retrieval framework that organizes large language models, biochemical computational tools and external knowledge resources into traceable evidence chains for bidirectional enzyme-to-reaction and reaction-to-enzyme retrieval. In four experimentally supported enzyme–reaction cases drawn from an E. coli metabolomics study, it ranked among 667 candidate proteins with a mean reciprocal rank of 0.833 using Qwen3.6-35B-A3B and 0.750 using GPT-5.6 Luna, above the general-purpose coding-agent combinations and domain methods listed in the comparison table. On public benchmarks it reached Top-1 query success rates of 82.3% on Enzyme-405 and 72.8% on Orphan-335.

  9. arXiv

    ElecLoad lets shifting event effects set the forecast segments: PJM WAPE falls from 4.31% to 3.69% when events pile up

    ElecLoad reads weather alerts, grid notices and news event descriptions alongside historical load, judging each event's relevance to the target load, the timing of its effect, and whether events reinforce or offset one another, then decides whether to forecast the whole horizon at once or split it where event effects change. On the segmented path, sub-agents draft each segment independently and the main agent reviews them and sends feedback to selected segments only. Across hourly load data including GEFCom2014, PJM and LCL it ranks near the top in most evaluation scenarios; under multiple events on PJM its WAPE is 3.69% versus 4.31% for a direct-only variant using the same model and inputs.

  10. arXiv

    SecureGARD puts a semantic gate on tool calls: zero observed attack success across three benchmarks and two backbones

    Tool-using LLM agents turn model output into real actions, making indirect prompt injection an execution-security problem. SecureGARD inserts an executable ontology between planning and tool execution, grounding heterogeneous tool calls into shared semantic actions and judging each as ready, waiting, or forbidden against the current task state. Under benign and indirect-injection settings on AgentDojo, ASB, and AgentDyn, the observed attack success rate is 0% on both backbones, with higher task utility under attack than the compared baselines.