aiqbench analysis
This research introduces SpectrumAudit, a novel framework for auditing wearable activity recognition models against adversarial perturbations that masquerade as sensor biases. By isolating persistent DC offsets from zero-mean variations, the study demonstrates how specific waveform injections can significantly degrade model robustness across diverse datasets and backbones. The findings highlight critical vulnerabilities in current HAR systems where seemingly smooth temporal signals exploit static sensor errors rather than genuine dynamic changes.
When Temporal Perturbations Act Like Sensor Biases: Label-Free Auditing of Wearable Activity Recognizers
arXiv:2609.29937v1 Announce Type: new Abstract: Wearable human-activity recognition (HAR) models operate across sensors, subjects, and backbones, yet a smooth waveform may appear temporal while exploiting a persistent sensor offset primarily. We introduce SpectrumAudit, a label-sealed audit that fits a phase-randomized full-window stimulus on calibration windows from subjects held out from training and testing. After selection, it replays its exact DC projection and budget-constrained zero-mean residual on the same frozen victim without refitting. Across 27 victims from three datasets and three backbones, the selected waveforms cause 2.87-40.83-point three-phase robust accuracy losses. Under this replay budget, DC is more damaging than AC on 24/27 victims and recovers at least 90% of the full drop on 22/27; all 5 failures occur on WISDM. In a held-out UTD-MHAD check, the selected waveform causes 13.49-pp accuracy and 11.68-pp macro-F1 losses, versus -0.66 pp for matched random changes. The audit diagnoses offset versus zero-mean variation under a common peak-budget cap. The code will be released upon acceptance.
Abstract:Wearable human-activity recognition (HAR) models operate across sensors, subjects, and backbones, yet a smooth waveform may appear temporal while exploiting a persistent sensor offset primarily. We introduce SpectrumAudit, a label-sealed audit that fits a phase-randomized full-window stimulus on calibration windows from subjects held out from training and testing. After selection, it replays its exact DC projection and budget-constrained zero-mean residual on the same frozen victim without refitting. Across 27 victims from three datasets and three backbones, the selected waveforms cause 2.87-40.83-point three-phase robust accuracy losses. Under this replay budget, DC is more damaging than AC on 24/27 victims and recovers at least 90% of the full drop on 22/27; all 5 failures occur on WISDM. In a held-out UTD-MHAD check, the selected waveform causes 13.49-pp accuracy and 11.68-pp macro-F1 losses, versus -0.66 pp for matched random changes. The audit diagnoses offset versus zero-mean variation under a common peak-budget cap. The code will be released upon acceptance.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
arXiv:2609.29937 [cs.LG]
(or arXiv:2609.29937v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.29937 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 257 words
Related coverage
PPA 7: Temporal Consistency and Texas Sharpshooting
LessWrong · Oct 14, 2026
Leakage-Free Evaluation and Distribution-Robust Spatio-Temporal Graph Learning for Inductive Kriging
arXiv · Aug 28, 2026
Reliability of LLM Judges for Evaluating Entity Alignment
arXiv · Oct 8, 2026
Presentation: Building Evals for AI Adoption: from Principles to Practice
InfoQ · May 29, 2026