Hefei University of Technology

2 related articles

Subscribe to updates

arXiv Oct 12, 2026
Skeleton action recognition gets a quantization plug-in: white-box attack success drops from near 100% to single digits, accuracy stays comparable
Mainstream contrastive models for self-supervised skeleton-based action recognition had almost no defense against adversarial attacks, where tiny perturbations to skeleton data could sharply degrade recognition. QCLR adds quantization as a plug-in module at two levels of the encoding pipeline, mapping continuous inputs and features to discrete codebooks so small perturbations are absorbed and gradient propagation is broken; integrated into SkeletonCLR, AimCLR and ActCLR on NTU-60, NTU-120 and PKU-MMD, it cuts PGD and SMART attack success rates to single digits while keeping classification accuracy comparable to the original models.
Mainstream contrastive models for self-supervised skeleton-based action recognition had almost no defense against adversarial attacks, where tiny perturbations to skeleton data could sharply degrade recognition. QCLR adds quantization as a plug-in module at two levels of the encoding pipeline, mapping continuous inputs and features to discrete codebooks so small perturbations are absorbed and gradient propagation is broken; integrated into SkeletonCLR, AimCLR and ActCLR on NTU-60, NTU-120 and PKU-MMD, it cuts PGD and SMART attack success rates to single digits while keeping classification accuracy comparable to the original models.
Central South University University of Leeds Hefei University of Technology
Nanzhu Chen（First author）
QCLR Skeleton-Based Human Action Recognition Contrastive Learning Vector Quantization Adversarial Robustness
arXiv Oct 1, 2026
OverAct benchmark shows all seven tool-calling agents exceed authorized scope, and SelfAudit cuts privacy-oriented excess by 43%
The work introduces the OverAct benchmark and a decision-theoretic framework to measure proactive over-authorization in structured tool-calling agents across eight privacy-sensitive domains, finding that all seven models from four families significantly exceed authorized scope, that request specificity is the strongest predictor of severity, that over-authorization grows sublinearly with tool-pool size, and that decoding temperature has little effect; it also proposes SelfAudit, a zero-shot inference-time method that generates request-grounded justifications and filters unjustified calls before execution, reducing privacy-oriented excess by 43% without oracle knowledge.
The work introduces the OverAct benchmark and a decision-theoretic framework to measure proactive over-authorization in structured tool-calling agents across eight privacy-sensitive domains, finding that all seven models from four families significantly exceed authorized scope, that request specificity is the strongest predictor of severity, that over-authorization grows sublinearly with tool-pool size, and that decoding temperature has little effect; it also proposes SelfAudit, a zero-shot inference-time method that generates request-grounded justifications and filters unjustified calls before execution, reducing privacy-oriented excess by 43% without oracle knowledge.
Alibaba Group (China)Hefei University of Technology East China Normal University
Taolin Zhang（First author）
OverAct SelfAudit Proactive over-authorization LLM Agents

You’re all caught up

Skeleton action recognition gets a quantization plug-in: white-box attack success drops from near 100% to single digits, accuracy stays comparable

OverAct benchmark shows all seven tool-calling agents exceed authorized scope, and SelfAudit cuts privacy-oriented excess by 43%