Skip to main content
Search

Ant Group (China)

16 related articles

Subscribe to updates
  1. arXiv

    An agent writes its execution experience into reviewable skills: DrugHarness designs and synthesizes an HPK1 compound with 14.19 nM cellular IC50

    Computational drug-discovery agents usually rebuild their tool calls for every task, so a successful strategy rarely carries over. DrugHarness places more than 80 scientific operations and biomedical resources in one workspace, induces action hypotheses with explicit applicability conditions from execution trajectories, and admits them as skills only after an independent review agent checks sources and runs targeted trials when needed. On MolBench, the configuration without evolution reaches 91.9% affinity-comparison accuracy, 10.8 percentage points above reported MolClaw-CC results; in retrospective evaluation, frozen learned skills raise Qwen's optimization success from 79.5% to 87.2%.

  2. arXiv

    No old samples, one autoencoder per class: TRACE cuts retained memory about 98% below TS-ACL

    For behavioral classes that arrive over time, TRACE stores no historical samples and never expands a shared classifier; it feeds a frozen temporal-spectral representation to a small autoencoder per class and labels by reconstruction error. Across five multivariate time-series class-incremental benchmarks it reaches 88.85% five-dataset average final accuracy with 9.17 MiB of retained memory, about 98.26% less memory than TS-ACL, the strongest analytical baseline by average accuracy, and an approximately 33.43% relative accuracy gain on GrabMyo.

  3. arXiv

    Omnimodal models mishear and missee without retraining: an inference-time nudge back to the relevant modality lifts CMM overall accuracy from 70.5% to 85.9%

    Omnimodal large language models answer wrongly when audio and video interfere with each other, yet the question-relevant unimodal branch often still holds the right answer; a training-free inference-time intervention steers the fused representation back toward that branch on conflicting samples, raising CMM overall accuracy from 70.5% to 85.9%.

  4. arXiv

    Phone agents fall to cross-modal injection: same payload in screenshot and UI text hits 85.4% attack success, single-channel repeats lag

    For multimodal-LLM phone agents, MobileInject places one malicious objective in both the screenshot and the interface text and runs 108 executable risk instances on a resettable Android emulator. Prior evaluations mostly watched a single text channel or judged only whether an attack succeeded; with exposure counts matched, one appearance per channel reached 85.4% attack success and 0% benign completion, while repeating within one channel reached only 72.9% or 27.8%. In 30 of 108 paired instances, neither single channel succeeded but the aligned pair did.

  5. arXiv

    Pruning calibration stops reading fixed text: decode-time activations lift Qwen3-32B writing score from 2.99 to 5.31

    LLM pruning now builds its calibration activations from the dense model's own autoregressive decode steps while discarding prefill, aligning the pruning objective with generation-time activations; it beats fixed-text calibration on WritingBench and ClassEval and reaches up to 1.48x end-to-end decoding speedup on A100 GPUs.

  6. arXiv

    UniSkill trains a skill proposer with contrastive action feedback, reaching 98.4% success on ALFWorld with stable joint training

    UniSkill lets a shared policy act as both the actor that executes environment actions and the proposer that emits skillbank edits (Add, Update, No Edit), and trains the proposer with contrastive action feedback that measures how replacing the retrieved skill changes the current actor's action log-likelihood gap between previously collected successful and failed trajectories, avoiding extra rollouts per proposal, with skill-edit support regularization preserving exploration; it reaches 98.4% success on ALFWorld and 84.7% success on WebShop while sustaining stable joint training over 250 steps.

  7. arXiv

    SF-MOPD pairs a fast student with a slow EMA reference to decouple capability interference in multi-teacher distillation, improving multimodal specialization while reducing general-capability degradation across model scales

    The work proposes Slow-Fast Multi-Teacher On-Policy Distillation (SF-MOPD), which couples a fast model updated directly by each teacher with a slow model that is an exponential moving average of the student, computes each teacher-induced update in log-probability space, and removes only the component pushing the fast model away from the slow model while retaining aligned and orthogonal components; experiments across multiple model scales show it mitigates capability interference, enhances specialized multimodal capabilities, and reduces average degradation on general-capability benchmarks, consistently outperforming vanilla MOPD.

  8. arXiv

    HDPO has LLMs list candidate solution outlines before selecting one to reason through, reportedly improving both reasoning and candidate-solution diversity

    The work proposes Hint-Guided Diversified Policy Optimization (HDPO), which lets a model first list all potential candidate solution outlines as hints and then select the most reliable one for further reasoning; the method has two stages, Cold Start for Structured Reasoning and Hint-Guided Diversified Reinforcement Learning, to incentivize diverse and reliable solutions along a "propose-select-think" trajectory, and experimental results are reported to show that HDPO effectively boosts LLM reasoning and enhances the diversity of candidate solutions as well as the model's ability to identify reliable solutions.

  9. arXiv

    UILoop recasts GUI reasoning as a cyclic Screen-UI elements-Action process and releases a 26K-sample UI Comprehension-Bench, reporting state-of-the-art UI understanding and superior GUI reasoning

    The work proposes the UI-in-the-Loop (UILoop) paradigm, which models GUI reasoning as a cyclic Screen-UI elements-Action process in which Multimodal Large Language Models explicitly learn the localization, semantic functions, and practical usage of key UI elements, enabling precise element discovery and interpretable reasoning; it further introduces a more challenging UI Comprehension task with three evaluation metrics and contributes a 26K-sample UI Comprehension-Bench, with experiments reporting state-of-the-art UI understanding and superior results on GUI reasoning tasks.

  10. arXiv

    Logits-to-Logic strengthens and filters last-layer logits to raise LLM logic consistency and reach state-of-the-art on multiple KGQA benchmarks

    Targeting the Logic Drift that appears in LLM outputs during structured knowledge reasoning, this work proposes the Logits-to-Logic framework, whose core modules are logits strengthening and logits filtering, to directly correct the logits produced in the autoregressive generation process; experiments report significantly improved logic consistency in structured knowledge reasoning and state-of-the-art performance on multiple KGQA benchmarks.