Implemented linear classifiers and neural networks first from scratch and then using Pytorch for medical image classification in Homework 1 for the Deep Learning course at IST (2023-24).

Friday, December 15, 2023.

This repository contains the implementation and solutions for Homework 1 in the Deep Learning course at IST for the academic year 2023-24. The assignment consists of multiple questions related to medical image classification using various machine learning and neural network techniques.

The project is organized as follows:

• download_octmnist.py: Script to download the OCTMNIST dataset.
• hw1-q1.py: Python skeleton code for Question 1.
• hw1-q2.py: PyTorch skeleton code for Question 2.
• README.md: Details about the project and instructions.
• Project.pdf: Project's assignement.

The homework comprises three main questions, each focusing on different aspects of deep learning:

1. Question 1: Linear classifiers and neural networks.
2. Question 2: Medical image classification with autodiff toolkit.
3. Question 3: Designing a multilayer perceptron for a Boolean function.

Each question has specific instructions and tasks to be completed. Here's a brief overview:

• Subtask (a): Implementing perceptron and reporting its performance.
• Subtask (b): Implementing logistic regression and comparing models with different learning rates.
• Subtask (c): Implementing a multi-layer perceptron without using neural network libraries.

• Subtask (1): Implementing logistic regression using stochastic gradient descent and tuning the learning rate.
• Subtask (2): Implementing a feed-forward neural network with dropout regularization and comparing different hyperparameters.

• Subtask (1): Analyzing the limitations of a single perceptron for a specific function.
• Subtask (2): Designing a multilayer perceptron to compute a Boolean function.