Online and continual learning using randomization based deep neural networks

• Main Author: Sreenivasan, Shiva
• Other Authors: Radhakrishnan K
• Format: Thesis-Master by Research
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/165774
• Institution: Nanyang Technological University

Deep neural networks have shown their promise in recent years with their state-of-the-art results. Yet, they suffer from some issues such as the time-consuming training process and catastrophic forgetting. In this work we look to overcome them by combining the advantages of an online learning process as new data arrives and a system with fast and effective learning capability such as the Random Vector Functional Link (RVFL) which is a Randomization based Deep Neural Network. Our approach involves allowing the model to grow incrementally as new data is made available so that it can more resemble real-life learning scenarios. Although RVFL network was proposed as a single-hidden layer feedforward neural networks (SLFNs), deep variants have been recently developed. As opposed to conventional neural networks adjusting network weights iteratively, RVFL uses a simple learning method without iterative parameter learning. Keywords: RVFL, Online Learning, Continual Learning.