This project is the final project of the academic course "Introduction to Machine Learning" (year 2024-2025).
The aim of the project is to develop a new type of learnable (parameterized in a way that backpropagation can optimize it) pooler (Pooling Operation, eg MaxPool2d) for standard CNN architecture.
The project is divided in 3 different parts:
- Literature review and existing poolers implementation in PyTorch
- Implementation of a new pooler in PyTorch
- Evaluation of the new pooler on CIFAR10 and MNIST using a standard CNN
The code will be develop in Python files while the notebook is the wrapper for the final product.
More about the project in the notebook.
Author : Francesco Bredariol SM3201379