Manuscript received October 7, 2023; revised December 27, 2023; accepted January 3, 2024; June 28, 2024

Abstract—Deep learning encompasses the inherent properties of scattered data and acquires a more abstract representation of data than conventional machine learning techniques. Nevertheless, existing deep learning algorithms perform inadequately on novel problems such as image classification as they generally require an extensive number of samples intended to train the model. One of the efficient methods to resolve the indicated drawback is meta-learning, acclaimed as learning to learn. Using antecedent knowledge to aid in the learning of new tasks improves meta-capacity for generalization to unfamiliar tasks. Meta-learning determines previous assignments intending to discover a representation that is easily adaptive to unknown challenges. Meta learning methodologies help find these components through multitudinous learning episodes by learning to solve a set of tasks instead of solving a single task at a time. Episodic meta-learning seeks to imitate a realistic environment by producing small artificial tasks from a substantial set of training tasks for meta-training and then moving on to the related method for meta-testing. The research is evaluated with meta learning algorithms like Prototypical Network and proto-Model-Agnostic Meta-Learning (MAML) with episodic meta learning on SVHN and Omniglot dataset reporting compelling enhancements on public benchmarks. In this research, the obtained results demonstrate a notable improvement and enhancement compared to existing methodologies, indicating a successful and impactful improvisation in the proposed methodology.

Keywords—meta learning, deep learning, few shot learning, prototypical networks proto-Model-Agnostic Meta-Learning (MAML)

Cite: Syeda Roohi Fatema and Sumana Maradithaya, "Meta Learning Approach Based on Episodic Learning for Few-Shot Image Classification," Journal of Image and Graphics, Vol. 12, No. 2, pp. 205-214, 2024.

Copyright © 2024 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.