Brain inspired lifelong learning model based on neural based learning classifier system for underwater data classification. (30th December 2021)

• Record Type: Journal Article
• Authors: Irfan, Muhammad; Jiangbin, Zheng; Iqbal, Muhammad; Masood, Zafar; Arif, Muhammad Hassan; Hassan, Syed Rauf ul
• Is Part Of: Expert systems with applications. Volume 186(2021)
• Publication Date: 2021-12-30
• Subjects: Lifelong learning -- Underwater image classification -- Learning classifier systems -- Convolutional neural network; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2021.115798
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: The general benchmark for success of an artificial intelligence system is its ability to imitate learning of the human brain. The human brain is capable of continuous learning over a lifespan. The learned knowledge is retained, augmented, fine-tuned and reused to perform new future tasks. At present, machine learning models perform well when carefully arranged, balanced and homogenized data is presented. However, most of these models undergo performance degradation when multiple tasks with incremental data are presented. Inspired by learning of the brain, in this study, we propose a lifelong learning model which extracts knowledge and utilizes the previously learned knowledge to solve the current problem. In the proposed model, firstly, we exploit various deep convolution blocks to extract non-trivial features from images, then a code fragment based learning classifier system with a rich knowledge encoding scheme is devised for knowledge extraction, transfer and reuse. We validate the proposed model with 2 incremental learning scenarios: (i) new instances (ii) new classes, on underwater synsets of the benchmark ImageNet dataset. Experiments results which are analyzed by using paired sampled statistical t-test, show that the proposed model outperforms baseline methods as well as deep convolution neural network based methods, with respect to classification accuracy. Highlights: Lifelong learning with knowledge extraction, retention and reuse capability. Exploitation of various CNN blocks for feature extraction. Model capable to cover both new instances and new classes scenarios. Investigation of multiple learning techniques. Better classification accuracy results as compared to deep CNN methods.