Agentic Anomaly Detection with ORCA-Style Dynamic Inductive Bias Adaptation in Multimodal Wearable Time Series Data

2026-08-09

AI Summary

This work addresses the challenge of modeling highly non-stationary multimodal physiological time-series data on resource-constrained wearable devices, where fixed temporal receptive fields struggle to adapt to heterogeneous signal patterns. To this end, the authors propose ORCA, a lightweight framework that dynamically selects discrete temporal receptive fields during inference based on real-time signal statistics, enabling state-dependent inductive bias adaptation without retraining. The key innovation lies in a proxy controller that requires neither additional trainable parameters nor auxiliary learning strategies, thereby eliminating reliance on predefined window sizes and incorporating a multi-scale temporal context selection mechanism. Experiments demonstrate that ORCA achieves an AUROC of 0.99 on a custom WBAN dataset—matching the performance of the best fixed-window methods—while exhibiting strong out-of-distribution generalization on MIMIC-IV.

Abstract

Wireless Body Area Networks (WBANs) generate multivariate physiological time series that are highly nonstationary and must often be processed under strict computational and memory constraints. A critical yet underexplored challenge in this setting is selecting an appropriate temporal receptive field, which serves as a strong inductive bias for anomaly detection models. Existing approaches typically rely on fixed temporal contexts, which can perform inconsistently across heterogeneous signal regimes and require dataset-specific tuning. We propose ORCA, an agentically controlled anomaly detection framework that dynamically adapts the temporal receptive field at inference time based on lightweight signal statistics. Rather than introducing additional trainable parameters or learned policies, ORCA employs a supervisory controller that autonomously selects among discrete temporal contexts, enabling state-dependent inductive bias adaptation without retraining. Across a custom WBAN dataset, ORCA achieves performance comparable to the strongest fixed-context baselines (AUROC = 0.99) while eliminating the need to tune temporal horizons in advance. We further evaluate ORCA on MIMIC-IV as a challenging out-of-distribution benchmark, observing conservative generalization behavior without performance collapse under heterogeneous clinical conditions. These results highlight adaptive temporal inductive bias control as a practical and robust design principle for anomaly detection in resource-constrained, nonstationary physiological time series.

Problem

• anomaly detection
• temporal receptive field
• inductive bias
• nonstationary time series
• wearable sensors

Innovation

• dynamic inductive bias
• temporal receptive field adaptation
• agentic anomaly detection
• nonstationary time series
• resource-constrained WBAN