gradient-checking

This repository contains the programming assignments and slides from the deep learning course from coursera offered by deeplearning.ai

A repository that contains all my work for deep learning specialization on coursera. This repository contains the programming assignments from the deep learning course from coursera offered by deeplearning.ai

Neural Network Deep learning specialization course offered via Coursera

Python Machine Learning Deep Neural Network implementation written from scratch with no dependencies. For fun and learning.

My Python solutions for the assignments in the machine learning class by andrew ng on coursera.

A five-course specialization covering the foundations of Deep Learning, from building CNNs, RNNs & LSTMs to choosing model configurations & paramaters like Adam, Dropout, BatchNorm, Xavier/He initialization, and others.

Improving Deep Neural Networks coding assignments

LoRA implemented from the paper math in NumPy with finite-difference gradient checks. Rank-2 hits 99.7% accuracy training 0.4% of params with 226x less optimizer memory; a paired experiment shows exactly when LoRA cannot work (no domain coverage: 0.66 vs 1.0).

A multilayer perceptron from scratch in pure numpy: forward pass, hand-derived backprop, SGD with momentum, trained on a non-linear spiral, with a gradient check proving the math (~1e-8) and an accuracy gate. No DL framework.

Computer vision workflow on CIFAR-10, from classical classifiers and NumPy neural networks to PyTorch CNNs, ResNet-18, and Vision Transformers.

Feed-forward neural network built from scratch in NumPy - He init, L2 regularization, and a finite-difference gradient check that verifies the analytical backprop to 1.6e-09

Python implementation of image classification models for CIFAR data, including linear classifiers, softmax stability checks, gradient checking, and a two-layer neural network.