Loading...

Continual Learning Algorithms Inspired by Human Learning

Banayeeanzadeh, Mohammad Amin | 2022

156 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 54846 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Soleymani Baghshah, Mahdieh
  7. Abstract:
  8. Despite the remarkable success of deep learning algorithms in recent years, it still has a long way to reach the status of human natural intelligence and to acquire the expected self-autonomy. As a result, many researchers in this field have focused on the development of these algorithms while taking inspiration from human cognitive behaviors. One of the disadvantages of current algorithms is the lack of their ability to learn in a continual manner while deployed in the environment. More precisely, deep learning models are not able to gradually gather knowledge from the environment and if they are in a situation of limited access to data, they will suffer from catastrophic forgetting; a phenomenon in which the model will lose performance on the data it has not observed for a while. One of the proposed methods to solve this problem is meta-continual learning in which the intelligent agent tries to address the forgetting problem by leveraging meta-learning techniques. In this dissertation, inspired by neuroscience and human learning behavior, we present a new approach to meta-continual learning. For this purpose, we will first go through a comparison between generative classifiers against discriminative classifiers and then explain their superiority in continual learning from both theoretical and practical aspects. In the next step, inspired again by neuroscience, we use the Bayesian approach and introduce a generative probabilistic classifier. Our experiments on Omniglot, Mini-Imagenet and CIFAR100 datasets show a 33, 32, and 23 percent difference between our method and the previous state-of-the-art. These results along with other intuitive experiments prove the efficiency and advantage of the proposed method in comparison with other meta-continual learning works
  9. Keywords:
  10. Deep Learning ; Cognitive Science ; Neuroscience ; Brain-Inspired Learning ; Continual Learning ; Meta-Continual Learning

 Digital Object List

 Bookmark

...see more