# LENET mechanism reference

Run `python examples/lenet.py` for deterministic self-checks or `python examples/lenet.py --json` for parity data. Python 3.9+; standard library only.

## Source and provenance

- Authors: Yann LeCun, Léon Bottou, Yoshua Bengio, Patrick Haffner.
- Version: 1998 author manuscript submitted to Proceedings of the IEEE.
- Primary source: https://bottou.org/papers/lecun-98h
- PDF: https://leon.bottou.org/publications/pdf/ieee-1998.pdf
- PDF SHA-256: `2b044160939a2a8ce6c95d7f3203e12a70287404aed9eca52cfed3fc1493aef2`
- Source crops: `public/images/papers/lenet/source-crops.json` records one-based PDF pages, PDF-point rectangles, render scale, dimensions and image hashes. These are attributed cropped/rasterized excerpts; no open license was verified.
- Reference authored for Abhik Sarkar's review, 2026-11-06. Outputs are calculated teaching values, not recorded checkpoint predictions.

## Scope and limitations

Models valid cross-correlation, scaled tanh, learned average/subsampling, original partial C3 connectivity and squared Euclidean RBF distances. The original sum-before-coefficient subsampling is represented equivalently by an average with a rescaled coefficient. Synthetic inputs and reduced-dimensional prototypes are authored illustrations. No trained LeNet, MNIST loader, loss, optimizer or accuracy reproduction is supplied.

## Reproducibility

`node --test test/lenet.test.mjs` checks independent mathematical invariants and JavaScript/Python parity. Canonical source lives in `examples/lenet.py`; named region excerpts and public downloads are regenerated by `scripts/sync-paper-examples.mjs`. Do not edit generated mirrors.

To verify crop bytes, obtain the pinned PDF and run `uv run --with pymupdf python scripts/render-paper-crops.py --manifest public/images/papers/lenet/source-crops.json --pdf /path/to/paper.pdf`. PyMuPDF is an authoring tool, not an application or reference dependency.
