01 / IMPLEMENTATION & CONTRIBUTION
What the work involves
Course implementations build dynamic computation graphs, broadcast-aware derivatives, topological backpropagation, parameterized layers, SGD, and a 784–256–128–10 classifier from NumPy.
Technical depth
Function/Context graph nodes; gradient accumulation along multiple paths; unbroadcasting; batched matmul derivatives; reductions and activation derivatives; numerically stabilized softmax cross-entropy; module/parameter discovery; dataset/batching; NumPy-only optimizer and train/test preprocessing.
The project family
uiuc/ece449/CS446_ECE449_SP2026_MP102 / RESULTS
What came out of it
An end-to-end teaching library spanning differentiation, batching, optimization, and MNIST training code, with supplied unit tests.
03 / SUPPORTING EVIDENCE
Follow the source
Implementation notes, project records, and supporting artifacts.
Source context & project scope
A CS446/ECE449 course study based on supplied scaffolding, not original autodiff research.
No Git history or explicit candidate-named report was found; public phrasing should retain the coursework attribution.
Do not claim generic PyTorch compatibility or production performance.
No measured classification accuracy is claimed.
Defines the supplied NumPy-only autograd assignment and starter boundaries.
SOURCE · 2026-09-17Implemented MLP, training-data normalization and SGD training loop.
SOURCE · 2026-09-17