SEDSP: A Salience-Driven 17-Moment Cognitive Engine for Ultra-Low Power Edge Intelligence

Read full text Summarise with VIRAT
Authors: Qiuhua Zhou, Kaiyao WU, Jiaxu ZHOU, Yifan WU
Cited by: 0
References: 27

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Abstract We present the Sparse Event-Driven Sequence Processing (SEDSP) engine, a bio-inspired neuromorphic architecture that translates the 17-moment Vithi cycle from Abhidhamma philosophy into a computationally efficient framework for edge intelligence. In contrast to traditional Transformers, which incur constant computational overhead regardless of input salience, SEDSP leverages Temporal Elasticity and Proactive Salience-Driven Gating to achieve true energy proportionality. Empirical evaluations demonstrate a 138× reduction in total energy consumption in sparse, real-world environments and a 10.4× improvement in active-inference efficiency compared to dense baselines. By employing a fixed 7-step recursive “Javana” loop within the 15-moment Cognitive Enhancement path, SEDSP recovers up to 21.6 percentage points of absolute accuracy in high-noise conditions (SNR = 5 dB) relative to single-pass feed-forward baselines, while localized “Karmic Registration” in the 17th moment effectively mitigates catastrophic forgetting during continual learning. These results position SEDSP as a promising solution to the Efficiency-Robustness Trilemma in ultra-low-power edge deployment.

Fields of Study & Subjects

Advanced Memory and Neural Computing Advanced Neural Network Applications IoT and Edge/Fog Computing

Publication details

Journal / source
Research Square
Publication date
25 March 2026
Language
EN
Document type
Preprint