Home // COGNITIVE 2015, The Seventh International Conference on Advanced Cognitive Technologies and Applications // View article

Adaptive Anomalies Detection with Deep Network

Authors:
Chao Wu
Yike Guo
Yajie Ma

Keywords: Cognitive sensing; deep learning; anomaly detection

Abstract:
In this paper, we try to apply inspirations from human cognition to design a more intelligent sensing and modeling system, which can adaptively detect anomalies. The target of intelligent sensing and modeling is not to get as much data as possible, or to build the most accurate model, but to establish an adaptive representation of sensing target and achieve balance between sensing performance requirement and system resource consumption. To achieve this goal, we adopt a working memory mechanism to facilitate the model to evolve with the target. We use a deep network with autoencoders as model representation, which is capable to model complex data with its nonlinear and hierarchical architecture. Since we typically only have partial observations from sensed target, we design a variance of autoencoder which can reconstruct corrupted input. We utilize attentional surprise mechanism to control model update. Training of the deep network is driven by surprises (which are also anomalies) detected (with data in working memory), which means model failure or target's new behavior. Due to partial observations, we are not able to minimize free-energy in a single round, but iteratively minimize it by keeping finding new optimization bounds. While both random and non-random sensor selection can create new optimization bounds, certain non-random methods like surprise minimization algorithm used in this paper demonstrate better performance. For evaluation, we conducted experiments on simulated data to test whether our methodology makes the model more adaptive, and got positive result. In the next step, we will try to apply the work on some real applications including ECG and EEG anomaly detection.

Pages: 181 to 186

Copyright: Copyright (c) IARIA, 2015

Publication date: March 22, 2015

Published in: conference

ISSN: 2308-4197

ISBN: 978-1-61208-390-2

Location: Nice, France

Dates: from March 22, 2015 to March 27, 2015