HUAWEI Noah's Ark Lab · 8 records

dateidtitleauthorsvenue↑
2026-11-192611.24102IDEA: Intent-Dominance Evidence Audit for Inference-Time Backdoor Detection in GUI AgentsZhizhuo Kou et al.arXiv4
2026-10-232610.30882Inherited, Not Invented: Tracing the Provenance of Wait in Reasoning ModelsZhizhuo Kou et al.arXiv16
2026-10-222610.21204Entertain, then Cancel: How Thinking Makes Reasoning Models Read Some and Or LiterallyZhizhuo Kou et al.arXiv14
2026-10-072610.05166A Safe Action Is Not Enough: Feasible-Future Decoding for Vision-Language-Action PoliciesTu Nguyen et al.arXiv3
2026-10-052609.38096Tail-Influence Sampling for CVaR Policy EvaluationPauline Bourigault et al.arXiv7
2026-06-092606.06735A Geometric Account of Activation Steering through Angle-Norm DecompositionGeorgii Aparin et al.arXiv26
2026-06-092606.07473Whisper Hallucination Detection and Mitigation via Hidden Representation Steering and Sparse AutoEncodersGeorgii Aparin et al.arXiv14
2026-06-042606.01804SpeechEditBench: A Bilingual Multi-Attribute Benchmark for Instruction-Guided Speech EditingHanlin Zhang et al.arXiv2