Skip to main content
Search

The Hong Kong University of Science and Technology (Guangzhou)

27 related articles

Subscribe to updates
  1. arXiv

    Offline RLVR stops supervising every token: GIST splits data by outcome and beats seven selectors on a shared pool

    Offline reinforcement learning with verifiable rewards usually supervises every token of every response in a fixed sampled pool, although responses and tokens are not equally informative. GIST splits the bidirectional KL mismatch between the policy and the Gibbs-optimal target of KL-regularized RLVR into a forward and a reverse term, then pairs correct and incorrect responses by sampling temperature and supervises only low-probability tokens of correct responses and high-probability tokens of incorrect ones. On Qwen3-0.6B/GSM8K with all methods choosing from the same candidates, it supervises 66.6% fewer tokens than training on all candidates and reaches 45.79 strict and 49.73 flexible accuracy, above seven selection baselines.

  2. arXiv

    Traffic lights skip the perception pipeline: Traffic-VL reads raw camera views and cuts average queues 10.4% over three weeks at ten intersections

    Traffic signal control has relied on structured traffic states from dedicated perception pipelines, or on converting simulator images into intermediate representations before deciding. Traffic-VL is an eight-billion-parameter vision-language model that maps synchronized multi-view images and natural-language phase descriptions directly to phase choices, trained with SynTra-V, which turns physics-based simulation into semantically audited photorealistic surveillance imagery. It matches the structured-state Traffic-R1 on five closed-loop benchmarks, reaches 82.1% phase-selection accuracy on real surveillance images, and over three weeks at ten intersections reduced average queue length by 10.4% relative to historical manual operation.

  3. arXiv

    Masked diffusion models lock answers too early: HiLo-RD distills large-window trajectories, lifting GSM8K accuracy by 9.4 points

    Masked diffusion language models can fix answer fragments before the supporting reasoning is resolved when the response window is small, and those fragments then become context for later predictions. HiLo-RD uses the same model's large-window trajectory as a frozen teacher and aligns its hidden representations to cropped small-window views at committed positions. At the same small inference budget, GSM8K accuracy rises from 68.84% to 78.24% while mean readable length falls from 362.8 to 271.3 tokens, and the gains extend to code generation without code-domain training.

  4. bioRxiv

    π-SPECFormer learns from 291 million mass spectra and flags six mutation peptides missing from databases in lung adenocarcinoma

    Conventional database searching assigns high-confidence peptides to only about 20% to 31% of acquired spectra, leaving the rest unread. π-SPECFormer was pre-trained on 291 million tandem mass spectra in two stages: self-supervised learning of fragmentation patterns, then refinement with peptide–spectrum match groupings. It beat task-specific baselines in de novo sequencing, spectrum clustering and modification detection, and in 103 lung adenocarcinoma samples it produced nine high-confidence mutation peptides, six of them absent from the public databases examined.

  5. arXiv

    SLA moves spatial alignment into continual pre-training: 83.8% average success across eight real deployment conditions, 24.0 points over the baseline

    SLA extends spatial alignment for vision-language-action models from task post-training into continual pre-training, and keeps it during downstream post-training. Initialized from a pretrained policy, it jointly optimizes action prediction and a cosine alignment between intermediate visual embeddings and spatial representations from a frozen 3D perception model, using feature-dependent gates to aggregate multiple VLA layers and probabilistic state dropout to reduce reliance on demonstration trajectories. It reaches 72.3% on the RoboTwin 2.0 Clean2Random setting and 83.8% average success across eight real deployment conditions, 24.0 percentage points above the baseline.

  6. arXiv

    Long-context RL training stops waiting on autoregressive rollouts: sparse self-drafting plus full-attention verification lifts throughput 1.39x

    Long-context RLVR rollout generation now drafts tokens with sparse attention and verifies them with full attention, raising end-to-end training throughput to 1.39x and 1.35x of autoregressive on Qwen3-1.7B and Qwen3-8B while keeping comparable task performance.

  7. arXiv

    Agent RL stops relying on its own luck: failed rollouts get corrected mid-sampling, lifting navigation pass rate from 35% to 57%

    Alibaba researchers introduce Rollout-Time Intervention, which corrects the decisive error step using reward-side rubrics whenever an entire rollout group fails, letting agent RL learn from successful trajectories the policy could not explore on its own and raising Navigation Pass@1 from GRPO's 35.33 to 57.33.

  8. bioRxiv

    Membrane interfaces without complex structures: sequence-only cross-family language model ensemble ranks interface residues at 0.643

    Predicting membrane-protein interaction interfaces has been limited by scarce experimental complexes and by sequence alignments that break down across hydrophobic bilayers. CHORUS combines three protein language model families, ESM-2, ESM-C and ProtT5, across layer depths and three downstream networks into a six-model ensemble that outputs per-residue interface probabilities from a single sequence, reaching a chain-macro average precision of 0.643 on 4,655 homology-isolated test chains, above the complete-coverage structure method ScanNet-MSA at 0.613. It also releases precomputed interface landscapes for 80,619 reviewed transmembrane proteins across 4,918 species.

  9. arXiv

    Grain growth without solving phase-field equations: VN law sets areas, a graph network rebuilds the microstructure, 137x faster on five benchmarks

    A two-dimensional grain-growth simulation hands area evolution to the von Neumann–Mullins law and microstructure reconstruction to a graph neural network, reaching the same coarsening stage on five initial microstructures in 38.43 s on average versus 5,247.01 s for phase-field.

  10. arXiv

    UniData turns a simple requirement into multi-round, nine-modality instructions, beating Self-Instruct and VIGC on reasonableness, clarity, detail and relevance

    The authors introduce UniData, a universal multimodal instruction generation pipeline that first expands user keywords into diverse events, then uses a LLaMA-3-backed multimodal instruction generator with image and music tokenizers to produce instructions round by round, correcting repetition and irrelevance at inference via inter-round correlation; they also build UniDataset with nine modalities, 20,000 entries and 17.5 rounds on average, and report that its data quality exceeds Self-Instruct and VIGC on all four GPT-guided metrics while fine-tuning on the generated data improves other multimodal models' understanding and generation.