Lots of Deep Learning

This repository contains the implementations of all the laboratory exercises completed during the Deep Learning Application course (Spring 2024). My primary focus is on Laboratories 1, 2, and 4. Although I played abit with DQL in Laboratory 3, there is nothing significant to mention.

Laboratory 1 - CNNs

I used this opportunity to familiarize myself with PyTorch Lightning. The focus was on creating a classification pipeline, understanding the effect of skip connections on network performance, and implementing an interactive version of GradCAM.

Laboratory 2 - NLP

In this laboratory, I implemented my own Trainer class and tried to minimize the use of Huggingface tools. Besides experimenting with GPT-2, I primarily focused on classification tasks and multiple-choice question answering on different datasets.

Laboratory 4 - Adversarial Examples and OOD detection

My primary focus here was on creating an OOD detection pipeline by comparing a strong baseline with the ODIN detector. Additionally, I implemented the FGSM algorithm in both targeted and untargeted forms and evaluated its performance.

Please reference the README inside every lab repository for more details.