Strictly Causal Streaming Video Anomaly Detection with a State-Space Core (arXiv:2608.24810)

Forum topic · 小凯 · 2026-08-27

Summary

A new paper on arXiv (2608.24810) by Yogesh Kumar introduces a strictly causal streaming video anomaly detector built on a Mamba-style state-space model. Unlike prior approaches that internally buffer clips or windows, the proposed method maintains a fixed-size state updated in O(1) time and memory per incoming frame, with no lookahead and no clip buffering, making it suitable for real-time edge deployment. The temporal core is a diagonal linear state-space recurrence equipped with an input- and state-dependent decay gate, trained self-supervised via causal next-embedding prediction on a frozen visual backbone. A key theoretical contribution is a closed-form relationship between the recurrence's decay spectrum, detection latency, and the shortest anomaly duration that can be reliably captured. The work also addresses the evaluation gap of prior methods, which benchmark efficiency only through GPU throughput rather than on the edge hardware they target. This combination of strict causality, constant-memory updates, and theoretical latency analysis offers a practically deployable framework for streaming anomaly detection on resource-constrained devices.

Paper Overview

Field: AI
Author: Yogesh Kumar
Published: 2026-08-25
arXiv: 2608.24810

Abstract (Translation)

Recent work has applied Mamba-style state-space models (SSMs) to video anomaly detection, yet existing approaches still rely on buffering clips or windows internally, lack a theoretical account of how temporal memory relates to detection latency, and benchmark efficiency only through GPU throughput rather than the edge hardware these methods are intended to target. The author introduces a strictly causal streaming anomaly detector whose fixed-size state is updated in O(1) time and memory per incoming frame, with no lookahead and no clip buffering.

Key Contributions

Strict causality: No lookahead, no clip buffering — the model is fully compatible with real-time streaming inference.
O(1) per-frame updates: A fixed-size state is updated in constant time and memory for every incoming frame, suitable for edge deployment.
State-space temporal core: A diagonal linear state-space recurrence with an input- and state-dependent decay gate.
Self-supervised training: Causal next-embedding prediction on a frozen visual backbone.
Theoretical grounding: A closed-form relationship is derived between the recurrence's decay spectrum, detection latency, and the shortest anomaly duration the model can reliably capture.
Realistic benchmarking: The paper argues that efficiency should be measured on edge hardware rather than GPU throughput alone.

Source Abstract (English, truncated in source)

> Recent work has applied Mamba style state space models (SSMs) to video anomaly detection, yet existing approaches still rely on buffering clips or windows internally, lack a theoretical account of how temporal memory relates to detection latency, and benchmark efficiency only through GPU throughput rather than the edge hardware these methods are intended to target. We introduce a strictly causal streaming anomaly detector whose fixed size state is updated in O(1) time and memory per incoming frame, with no lookahead and no clip buffering. Its temporal core is a diagonal linear state space recurrence with an input and state dependent decay gate, trained self supervised through causal next embedding prediction on a frozen visual backbone. We derive a closed form relationship between the recu...

---

*Automatically collected on 2026-08-27.*