https://github.com/lancelet/digit-from-scratch

MNIST digit classification using a LeNet5 clone
https://github.com/lancelet/digit-from-scratch

Last synced: 3 months ago
JSON representation

MNIST digit classification using a LeNet5 clone

• Host: GitHub
• URL: https://github.com/lancelet/digit-from-scratch
• Owner: lancelet
• License: bsd-3-clause
• Created: 2021-01-13T06:51:49.000Z (over 4 years ago)
• Default Branch: main
• Last Pushed: 2021-02-02T01:48:02.000Z (over 4 years ago)
• Last Synced: 2025-01-13T21:27:17.283Z (4 months ago)
• Language: Haskell
• Size: 38.1 KB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • Changelog: CHANGELOG.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Learning Project: LeNet 5 Digit Classification

[![haskell.yml](https://github.com/lancelet/digit-from-scratch/workflows/Haskell%20CI/badge.svg)](https://github.com/lancelet/digit-from-scratch/actions?query=workflow%3A%22Haskell+CI%22)

Working through a re-creation of LeNet 5 from scratch, without using existing

deep learning packages.

Data and paper link: http://yann.lecun.com/exdb/mnist/

Build:

``` sh

./download-mnist.sh  # download the MNIST images to ./mnist-data

cabal run

```

## Progress so far

Work-in-progress:

  - [x] Loading MNIST images and label sets from raw binary

  - [x] Convolution layer forward evaluation

  - [x] Convolution layer back-propagation

  - [x] CI

  - [ ] Numerical test of gradient for convolution layer

  - [ ] Sigmoidal activation function

  - [ ] Sub-sampling layer forward evaluation

  - [ ] Sub-sampling layer back-propagation

  - [ ] Fully-connected layer forward evaluation

  - [ ] Fully-connected layer back-propagation

  - [ ] RBF units