https://github.com/franalgaba/neural-network-cairo

Neural Network implementation from scratch for MNIST using Cairo 1.0
https://github.com/franalgaba/neural-network-cairo

cairo-lang machine-learning starknet

Last synced: about 2 months ago
JSON representation

Neural Network implementation from scratch for MNIST using Cairo 1.0

• Host: GitHub
• URL: https://github.com/franalgaba/neural-network-cairo
• Owner: franalgaba
• License: mit
• Created: 2023-03-04T15:56:56.000Z (almost 2 years ago)
• Default Branch: main
• Last Pushed: 2023-04-01T15:55:53.000Z (over 1 year ago)
• Last Synced: 2024-07-30T20:34:35.353Z (4 months ago)
• Topics: cairo-lang, machine-learning, starknet
• Language: Rust
• Homepage:
• Size: 8.78 MB
• Stars: 9
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-cairo - `neural-network-cairo`
• awesome-starknet - neural-network-cairo - Neural Network implementation from scratch for MNIST. (Additional developer resources)
• awesome-rust-list - franalgaba/neural-network-cairo - network-cairo?style=social"/> : Neural Network implementation from scratch for MNIST using Cairo 1.0. (Machine Learning)
• awesome-rust-list - franalgaba/neural-network-cairo - network-cairo?style=social"/> : Neural Network implementation from scratch for MNIST using Cairo 1.0. (Web3 and ZKP Framework)

README

        


  

    

  



# Neural Network for MNIST in Cairo 1.0

Implementation of a Neural Network from scratch using Cairo 1.0 for MNIST predictions.

The NN has a simple two-layer architecture: 

  - Input layer 𝑎[0] will have 784 units corresponding to the 784 pixels in each 28x28 input image. 

  - A hidden layer 𝑎[1] will have 10 units with ReLU activation.

  - Output layer 𝑎[2] will have 10 units corresponding to the ten digit classes with softmax activation.

Functionalities implemented in Cairo 1.0:

 - [X] Vector implementation with operations: sum, max, min, argmax.

 - [X] Matrix implementation with operations: get, dot, add, len.

 - [X] Tensor implementation.

 - [X] 8-bit weight quantization based in [ONNX quantization](https://onnxruntime.ai/docs/performance/quantization.html#quantization-overview).

 - [X] ReLU activation.

 - [X] Forward propagation of NN.

 - [X] Predict method for NN.

 - [X] [Pseudo-softmax activation](https://www.nature.com/articles/s41598-021-94691-7) optimized for quantized values.

 - [X] Weight loading into Cairo NN from trained Tensorflow NN.

 - [X] MNIST inferences using Cairo NN.

[Built with **`auditless/cairo-template`**](https://github.com/auditless/cairo-template)

## Working with the project

Currently supports building and testing contracts.

### Build

Build the contracts.

```bash

$ make build

```

### Test

Run the tests in `src/test`:

```bash

$ make test

```

### Format

Format the Cairo source code (using Scarb):

```bash

$ make fmt

```

### Credits

- [cairo-ml](https://github.com/raphaelDkhn/cairo_ml) by [Raphael Doukhan](https://twitter.com/raphael_dkhn).