Skip to main content
Search

Beihang University

28 related articles

Subscribe to updates
  1. arXiv

    Candidate trajectory scoring drops per-candidate rollouts: RFWorld reads a shared scene memory and reaches 90.5 PDMS on NAVSIM

    End-to-end driving must pick one of several candidate trajectories, and earlier world-model evaluators explicitly predicted future states for each candidate over the planning horizon, so inference cost grew with candidates and horizon while prediction errors accumulated. RFWorld instead builds one trajectory-agnostic shared scene memory, shapes it during training with a lightweight temporal readout under semantic supervision, and lets candidates query it through sparse spatial readout at inference. On NAVSIM-v1 with a ResNet-34 backbone, RFWorld reaches 90.5 PDMS, above DriveDPO at 90.0 and ResWorld at 89.0 in the reported table.

  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

    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.

  4. arXiv

    A shared target for cross-lingual jailbreaks: a safety kernel of 0.12% of neurons lets masked fine-tuning cut attack success in many languages

    Large language models refuse the same harmful request in one language but may comply when it is rephrased in another. Researchers first identified safety-related neurons independently within each language, then took the intersection of the candidate sets from six languages to obtain a cross-lingually shared "safety kernel," and tested its functional role with activation suppression and activation-state transfer. Building on it, SK-Masked LoRA restricts fine-tuning updates to the feed-forward network output dimensions corresponding to the safety kernel, lowering attack success rates across several models and multilingual safety benchmarks, with some effect on languages excluded from kernel discovery.

  5. arXiv

    WebCoder trains frontend agents on a living project: 72.8 requirement satisfaction beats Claude-Opus-4.8's 65.1

    WebCoder-27B, trained during an internship at Zhipu AI, keeps reproduction, generation, modification and repair on one evolving project state and scores reinforcement learning by how well the executable application meets task requirements. On the new Webdev-Bench its requirement satisfaction rises from 64.8 for the Qwen3.8-27B base model to 72.8, above Claude-Opus-4.8 at 65.1; on Vision2Web its average visual and functional score rises from 41.1 to 62.4, on par with Claude-Opus-4.8 at 62.5.

  6. arXiv

    GRPO safety alignment loses its signal on hard prompts: Beacon restores a safe direction and cuts average safety error by up to 32.83 points across four models

    Safety alignment of large reasoning models commonly relies on GRPO, but responses sampled for one prompt often share the same safety outcome, so the within-group relative advantage vanishes and the model gets no direction toward safety on persistently unsafe prompts. Beacon pairs safety signal recovery, which filters high-contrast trajectories and directly supervises a safe reference response when the signal collapses, with safety-utility decoupled alignment, which assigns category-aware rewards and applies KL anchoring only to benign prompts. Across four Qwen3 and DeepSeek-R1-Distill variants and eight benchmarks, average safety error falls by 17.69 to 32.83 percentage points relative to the initial policies, with XSTest over-refusal between 0.40% and 1.20%.

  7. arXiv

    Interactive video world models treat old frames as false: re-dating by valid time lifts the current-state share from 0.36 to 0.62 after an undo

    Interactive video world models draw frames from a memory bank stamped with capture times, and a frozen generator reads the newest timestamp as truth, so after an agent undoes an action the only view of the present is an older frame judged outdated. ReDate keeps every frame and re-dates retrieved tokens by valid time read from the action log, so a reverted region reads as one that has just reached a new state, with no training. On FAR's 352 reverted steps the current-state share rises from 0.36 to 0.62 and the outdated share falls from 0.27 to 0.09; on HY-WorldPlay it rises from 0.003 to 0.19.

  8. arXiv

    Similar agent skills stop being judged by description alone: a capability graph scores the weakest link and lifts ALFWorld success from 22% to 29%

    As skill libraries grow, semantically similar skills become hard to tell apart from descriptions alone. SKALOS builds a multi-hierarchy capability graph of capability, method, artifact and constraint nodes, uses Max–Min witness learning to require that one implementation jointly covers every query requirement, and calibrates extraction errors with ranking feedback. On SkillRet its NDCG@10 is 2.68 points above a matched text-only reranker; on controlled ALFWorld tasks, witness-guided skill selection raises success from 22% to 29%.

  9. arXiv

    Private vector search stops scaling with database size: HNSecW brings online ANN queries to logarithmic time while holding recall at billion scale

    In the symmetric-privacy setting where both the query and the database must stay hidden, HNSecW turns HNSW graph-index approximate nearest neighbor search into a fully data-independent procedure that runs under secure multi-party computation, giving single-query online communication and local computation logarithmic complexity. Earlier protocols of this kind mostly used cluster-based pipelines whose online communication grows linearly with database size. On million- to billion-scale benchmarks, HNSecW keeps recall comparable to plaintext HNSW and reaches up to orders-of-magnitude end-to-end speedup over 2PC PANTHER on a wide-area network.

  10. Vehicles Leave After One-Shot Training: Exact Federated Editing Cuts Cumulative Time 36%–44% in Mobility Simulation

    One-shot federated learning finishes training in a single communication round, but the resulting model is frozen once devices join or leave, leaving full retraining as the fallback. F-SOSA draws a fixed-size uniform random client subset so outputs under neighboring device sets stay bounded in total variation, then rewrites curvature-aware fusion as a projection over client-additive sufficient statistics, enabling F-SOSA-U for withdrawing a contribution and F-SOSA-I for adding one. In a SUMO urban mobility simulation, both match retraining accuracy while cutting cumulative time and communication by 36%–44% and 22%–33%.