Pytorch Playground for Mathematical Reasoning Dataset

Neural Network build with numpy and math (no pytorch, keras etc.) for the MNIST dataset

Convolutional Neural Network build with numpy and math (no pytorch, keras etc.) for the MNIST dataset

No tf, no pytorch, just math using MNIST dataset

Neural Net From Scratch with Numpy and Math (No PyTorch, TensorFlow etc...) on the Fashion MNIST dataset

Neural ODE classifier in PyTorch using the adjoint sensitivity method for memory-efficient backprop through a continuous-depth dopri5 ODE solver. Trained on a spiral dataset vs ResNet baseline — includes phase portraits, decision boundaries, and inline adjoint math commentary.

This project fine-tunes the LLaMA-2-7B model on the GSM8K dataset to enhance mathematical reasoning. It utilizes Hugging Face Transformers, PyTorch, and PEFT techniques like LoRA for efficient adaptation, optimizing the model for solving complex math problems.

Implemented back-propagation from scratch in Python, focusing on basic operations (addition, multiplication, power, ReLU) and manual gradient computation for the MNIST dataset. This project, utilizing only math and numpy, aimed to deepen my understanding of neural networks' core algorithms without relying on frameworks like PyTorch or TensorFlow.

This project showcases a neural network built entirely from scratch using only mathematical principles and Python. No TensorFlow, no PyTorch—just pure math and code. The goal is to achieve high accuracy in digit classification using the MNIST dataset while deepening the understanding of neural networks' underlying mechanics.

This repo built neural networks completely from scratch using just NumPy and trained them on the MNIST dataset from Samson Zhang. The goal was to deeply understand how neural networks work under the hood — from forward and backward propagation to gradient descent and weight updates. No TensorFlow or PyTorch here, just pure math and code!