University of Pisa, UNIPI
Computer Science Department
Authors: Irene Pisani
July, 2023

Continual Learning

Final project for CL course (Prof. Antonio Carta)

An implementation of Model Agnostic Meta Learning for few shot supervised image classification

• Dataset: Omniglot dataset
• Algorithm : MAML_Algorithm.ipynb
• Experimental results: MAML_ExperimentalResults.ipynb
• Presentation: MAML_Slides.pdf

Project objective

1. Provide an implementation of MAML for few shot supervised learning using Pytorch.
2. Reproduce the experiments on the Omniglot dataset. Follow the original experimental protocol of N-way classification with 1 or 5 shots and N equal to 5 or 20.
3. Performance comparison between the obtained performance and original ones.
4. Analysis of the impact of the number of inner SGD steps during training and evaluation.