Skip to main content
Search

Nanjing University

34 related articles

Subscribe to updates
  1. arXiv

    Calpha lets a general chat model decide when to fold: mean TM-score 0.808 on single chains, still behind the direct baseline on long chains

    Calpha wires protein folding into a general chat model: Qwen3.5-9B reads a sequence with ESM2 residue features, emits a <structure> token when asked to fold, and conditions a frozen SimpleFold-3B head on its own residue hidden states to produce all-atom coordinates. Under a fixed folding request this configuration reaches mean TM-scores of 0.808 on FoldBench-334, 0.635 on CASP16 domains and 0.900 on a CAMEO-style PDB set, but on 300–600-residue chains the mean paired differences are −0.044, −0.160 and −0.025, all below the direct ESM2→SimpleFold-3B baseline.

  2. arXiv

    Crystal sequences stop following coordinates: topology-guided ordering lifts single-shot structure-prediction match from 56.10% to 61.92% at the same model size

    Crystal generative models write three-dimensional periodic structures as one-dimensional sequences, but ordering atoms by real-space coordinates lets small geometric perturbations reshuffle the sequence and inject variability unrelated to connectivity. CGTransformer instead orders atoms by periodic quotient-graph topology, combining element, symmetry and graph-embedded coordinates. On MP-20, the same 86M model raises single-shot structure-prediction match from 56.10% under coordinate ordering to 61.92%, and in conditional generation of 3.0–3.5 eV wide-bandgap semiconductors it increases the share of stable, unique and novel candidates.

  3. arXiv

    Chain-of-thought solves unseen multi-step tasks by reusing one step predictor, with error growing sublinearly in depth

    Chain-of-thought out-of-distribution generalization is framed as compositional generalization: training supplies only single-step atomic transitions, while the target task requires composing many steps. Theory bounds target risk by the sum of step risks times a depth-independent coefficient, and shows that sharing one predictor makes the estimation term shrink as the inverse square root of depth rather than grow linearly. Atom-Task and GSM8K experiments support short-to-long and unseen-combination transfer, and show explicit state tracking is essential.

  4. arXiv

    Streaming video-to-audio closes the alignment gap: LiveFoley needs only 0.08 s of frontend look-ahead and runs at 156 ms end-to-end

    Real-time video-to-audio has been held back by two bottlenecks: visual frontends shrink their receptive fields to cut look-ahead, and causal diffusion backends denoise each patch in isolation, so both alignment and audio quality trail offline models. LiveFoley uses a trailing sliding window to turn off-the-shelf non-causal segment-level encoders into streaming modules without retraining, cutting frontend look-ahead to 0.08 s, and widens the backend denoising window toward history. On the VGGSound test set, LiveFoley-48k reaches a DeSync of 0.48, on par with offline MMAudio-L at 0.47, with 156 ms end-to-end latency on a single A800 GPU.

  5. bioRxiv

    Protein language model embeddings already encode N-glycosylation sites: MaskNGly reaches AUC 0.940 on N-GlyDE without training

    MaskNGly treats protein language model embeddings as a readable source of coevolutionary signal, using a mask-and-probe readout to decide directly whether an N-X-S/T sequon will be N-glycosylated, without training on annotated data. On the N-GlyDE test set it reaches an AUC of 0.940 and an MCC of 0.808, comparable to published supervised neural-network predictors; in independent glycoproteomic data, 106 of 111 previously unreported sites scored above the 0.5 threshold.

  6. 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%.

  7. arXiv

    RSI-Env lets an agent rewrite both model and harness, then hands the next round to the accepted version, with gains even without benchmark feedback

    RSI-Env places model training and harness modification in one persistent sandbox, with an independent Host Controller handling services, validity checks, versioning and capability evaluation, and in Self mode each accepted model–harness version takes over the next improvement round. Starting from Qwen3.8-27B with a minimal harness, Open-General Self, which receives no benchmark feedback, raises GPQA Diamond accuracy from 65.2% to 73.2% and AIME 2026 from 53.3% to 56.7%; Knowledge Self lifts SimpleQA from 82.0% to 85.0% and then 88.0%, while the same runs fall back on other benchmarks.

  8. arXiv

    Visual tokens need not be ranked alike: BtsPruner keeps 32 of 576 and retains 91.6% of the dense model's performance

    Large vision-language models process hundreds to thousands of visual tokens per image, and inference cost rises with them. Earlier training-free pruning methods ranked all visual tokens by one importance criterion, overlooking that tokens take on different roles as encoding proceeds. BtsPruner locates the role transition from changes in attention concentration, then picks globally referenced representative tokens from later layers and recursively retrieves local detail tokens from earlier layers, preserving both roles under a fixed budget; on LLaVA-1.5-7B it keeps only 32 of 576 visual tokens (5.6%) and retains 91.6% of the dense model's performance.

  9. aclanthology.org

    Science rules move from the prompt into decoding: layered constraints lift average accuracy 12% across three scientific tasks

    Large language models still fabricate reasoning that contradicts domain rules in specialized scientific tasks. SciDC has a general model compile natural-language scientific knowledge into three layers of rules, top-level reasoning steps, middle-level conditional checks and bottom-level token masks, which a locally deployed domain model enforces during decoding. Across industrial formulation design, thyroid cancer TNM staging and single-step retrosynthesis, average accuracy rises 12% over vanilla generation, and ablations show removing the top or bottom layer sharply lowers formulation accuracy.

  10. 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.