GEMM-SaFIN(FRIE)++: explainable artificial intelligence visualisation system with episodic memory

Ko, N. M., Xie, C., Cao, Q. and Quek, C. (2023) GEMM-SaFIN(FRIE)++: explainable artificial intelligence visualisation system with episodic memory. In: Daimi, K., Alsadoon, A. and Coelho, L. (eds.) Cutting Edge Applications of Computational Intelligence Tools and Techniques. Series: Studies in Computational Intelligence (1118). Springer, pp. 255-284. ISBN 9783031441264 (doi: 10.1007/978-3-031-44127-1_12)

Abstract

Neuro-fuzzy network systems take advantage on functionalities of hybrid fuzzy logic and neural networks approaches. IF–THEN fuzzy rules allow good interpretability for human experts to understand the correlation between inputs and outputs. However, only designers know the mechanism and behavior of neuro-fuzzy systems. Details on how a neuro-fuzzy system derives and formulates the predictions or rules are unknown to users. It is due to the lack of transparency of the design and connections of neuro-fuzzy systems. We propose a novel explainable artificial intelligent (AI) visualization system for the neuro-fuzzy architecture, named general episodic memory mechanism (GEMM) Self Adaptive Fuzzy Inference Network with Fuzzy Rule Interpolation or Extrapolation with online learning capabilities (i.e., GEMM-SaFIN(FRIE)++). The proposed explainable AI visualization system is designed in a form of graphical user interface to assist users better understanding inner function mechanism on how rules are generated and how conclusions are drawn from the data fed into the neuro-fuzzy system. One of the challenges for neuro-fuzzy systems is making real-time predictions in the fast changing applications where data can be sparse. It may not be able to automatically detect and react to the occurrence of concept drifts, affecting the online learning capabilities. GEMM-SaFIN(FRIE)++ employs fuzzy rule interpolation and extrapolation techniques to make inference when concept drifts are detected. The GEMM mimicking human brains is employed to capture and retrieve from past events that GEMM-SaFIN(FRIE)++ learns. This is done by storing and retrieving them from an episodic memory storage during the transient event behaviors. Several experiments are conducted to evaluate the performance of the GEMM-SaFIN(FRIE)++. Firstly, three sub datasets in Nakanishi dataset including a sub dataset for daily price of stock in a stock market are utilised in the experiments. Secondly, GEMM-SaFIN(FRIE)++ is used to detect market trend reversal for two stock market indexes: Sand P 500 and DJIA index, under various events through the period during COVID-19, subprime and 9–11 attack. Encouraging experiment results are observed for event detections to capture a shift in stock prices.

• Item Type: Book Sections
• Additional Information: eISBN: 9783031441271.
• Status: Published
• Glasgow Author(s) Enlighten ID: Cao, Dr Qi
• Authors: Ko, N. M., Xie, C., Cao, Q., and Quek, C.
• College/School: College of Science and Engineering > School of Computing Science
• Publisher: Springer
• ISBN: 9783031441264