Deep Learning Programming Assignments (Jan–Apr 2025)

This repository contains solutions to Deep Learning lab assignments, each in a separate folder named LAB1, LAB2, ..., LAB11, corresponding to the assignments listed below.

---

📚 Lab Index

LAB1 - Feed Forward & Back-Propagation Learning Algorithm

• Build a perceptron and multi-layer neural network from scratch.
• Dataset: Iris Dataset

LAB2 - ANN for MNIST Digit Classification

• Fully connected neural network for classifying handwritten digits.
• Dataset: MNIST Dataset

LAB3 - CNN for MNIST Digit Classification

• Convolutional neural network model for MNIST.
• Includes comparison with FCNN (from LAB2).
• Dataset: MNIST Dataset

LAB4 - Binary Classification of Skin Lesions (ResNet-152)

• Implemented ResNet from scratch using basic PyTorch/TensorFlow layers.
• Dataset: ISIC 2020 Skin Lesion Dataset

LAB5 - Autoencoders

• MNIST dimensionality reduction using PCA and Autoencoder.
• Anomaly detection using Autoencoder (Melanoma dataset).
• Dataset : Melanoma dataset (split from LAB4)

LAB6 - Variational Autoencoders (VAE)

• VAE on FashionMNIST for synthetic image generation.
• Dataset: FashionMNIST

LAB7 - DeepFake Generation using GAN

• GAN trained on CelebA dataset for face generation.
• Dataset: CelebA Dataset
• Main source of the dataset: CelebA Dataset
• PyTorch: CelebA Dataset
• Tensorflow: CelebA Dataset

LAB8 - Sentiment Classification using RNN & LSTM

• RNN and LSTM for classifying movie reviews.
• Dataset: Stanford Sentiment Treebank v2 (SST-2)

LAB9 - Brain Tumor Segmentation using U-Net

• U-Net model for medical image segmentation.
• Dataset: LGG MRI Segmentation (Kaggle)

LAB10 - News Summarization using Transformers

• Fine-tuning BART model on news dataset.
• Dataset: ILSUM English Subset

LAB11 - Image Classification using ViT

• Classification of chest X-ray images into two classes: No Finding and With Findings using a Vision Transformer (ViT).

---

📌 Notes

• All notebooks are named as per the lab (e.g., CS22Bxxxx_lab03.ipynb)
• Report files (if required) are named consistently and included in relevant folders.