What's inside the box? Building a deep learning framework from scratch.
summary
Builds a minimalist deep learning framework in Python in under 90 minutes that uses a NumPy backend, implements elementwise ops, matrix multiply, ReLU, reduction, a Tensor container with autograd engine, Module/Parameter bookkeeping and a simple SGD optimizer. Trains a small MLP on synthetic data (1,024 samples, 10 features) and shows loss decreasing, notes no GPU/primitives or speed optimizations and provides a template plus solutions repo.
AI-written summary · about
chapters (10)
- 0:00Building a DL framework
- 1:34Why build from scratch
- 19:26Setting up the project and prerequisites
- 29:09Implementing multiplication
- 40:45Tensor class and autograd
- 50:15Module class for layers
- 1:00:23Training loop and loss
- 1:16:40Training and troubleshooting
- 1:17:22Primitives and backends
- 1:18:47Q&A
related talks
30mAnyone can Build Great Deep Learning Applications - Deep Numpy
35mDeconstructing Deep Learning
01h 10mDemystifying Deep Learning for Data Scientists
01h 26mDeep Neural Networks with PyTorch
01h 34mDeep Learning Hands-on Workshop
43mA Gentle Introduction to Neural Networks
43m(Deep) Learn You a Neural Net for Great Good!
20mLearn You a PyTorch!