Pablo Kvitca
On This Page

Minitorch

Archived

A from-scratch implementation of a PyTorch-like deep learning library in Python, completed across three modules for the DS4440 course at Northeastern University (Fall 2020).

Python

Overview

A ground-up implementation of a miniature deep learning library inspired by PyTorch, built across three progressive modules for DS4440 (Practical Neural Networks) at Northeastern University. Private repositories, class project.

Modules

  • Module 0 — Fundamentals: basic operators, testing infrastructure, and the core module system
  • Module 1 — Autodiff: automatic differentiation engine with scalar and tensor support
  • Module 2 — Tensors: tensor operations, broadcasting, and efficient computation

Each module builds on the previous one, culminating in a working library capable of training simple neural networks from scratch.