A collection of machine learning concepts and implementations I've explored. Each topic has been moved to its own dedicated repository for better organization and focused development.
An interpretability technique for deep neural networks that attributes predictions to input features by integrating gradients along a path from a baseline to the actual input. This implementation covers concepts from basic image gradients to integrated gradients, demonstrating how to identify which features most influence model decisions.
A from-scratch implementation of neural networks following Sentdex's "Neural Networks from Scratch" course. Built without external ML libraries, this repository covers fundamental concepts including neurons, layers, activation functions, softmax, and loss calculations, providing deep insights into the mathematical foundations of neural networks.
Exploration of optimal transport theory and its applications in machine learning. Includes implementations of the Wasserstein metric and practical applications like color transfer between images, demonstrating how optimal transport provides powerful tools for comparing probability distributions.
A semantic matching algorithm that ranks text entries based on embedding similarity using sentence transformers. Originally developed as the core algorithm for the Up2Mates project, it uses cosine similarity on pre-trained transformer embeddings to provide instant, relevance-based sorting of user submissions, demonstrating practical applications of semantic search in real-world matching systems.