Evaluation of multiple graph neural network models—GCN, GAT, GraphSAGE, MPNN and DGI—for node classification on graph-structured data. Preprocessing includes feature normalization and adjacency-matrix regularization, and an ensemble of model predictions boosts performance. The best ensemble achieves 83.47% test accuracy.

A PyTorch-based Deep Q-Network (DQN) implementation to solve the LunarLander-v3 environment using Gymnasium. Includes custom neural network design, experience replay, agent training, and performance visualization.

Implementations of neural networks in python for the classification of MNIST datasets.

This project outlines from scratch implementation of softmax regression which utilises softmax function to even out probabilities across k classes and cross entropy loss as loss function which were purely coded on NumPy and some Pandas

Assignment 1 : To build a neural network using softmax as activation function

JavaFx Application for Convolutional Network to perfom Image Classification using Softmax Output Layer, Back Propagation, Gradient Descent, Partial Derivatives, Matrix Flattening, Matrix Unfolding, Concurrent Task, Performance Histogram, Confusion Matrix

The implementation of Coordinate Descent Method Accelerated by Universal Metaalgorithm with efficient amortised complexity of iteration & Experiments with sparse SoftMax function, where the proposed method is better than FGM