How do I train a neural network on an ESP32 using MicroMLP?

Import the library, create a network with MicroMLP.Create(), add examples using AddExample(), and train with LearnExamples(). Ensure your microcontroller has sufficient memory for the network and data, as shown in the XOR example.

MicroMLP vs TensorFlow Lite for Microcontrollers: which should I use?

MicroMLP is simpler and lighter, ideal for basic MLP tasks on MicroPython. TensorFlow Lite offers more features and pre-trained models but requires more resources and is in C/C++. Choose based on project complexity and platform constraints.

What activation functions does MicroMLP support?

It supports Heaviside, Sigmoid, TanH, SoftPlus, ReLU, and Gaussian activation functions, which can be configured per layer for flexible network design, as listed in the README tables.

How to save and load a trained model in MicroMLP?

Use SaveToFile(filename) to export the network to a JSON file, and LoadFromFile(filename) to reload it. This allows easy deployment and reuse of models without retraining.

Can MicroMLP be used for real-time inference on embedded devices?

Yes, its lightweight nature enables fast inference on microcontrollers like ESP32, but training might be slow due to limited processing power. Optimize by adjusting parameters like Eta and Alpha.

Does MicroMLP work with image data or only simple inputs?

It can handle flattened image data via MLP, but it's not optimized for image recognition since it lacks convolutional layers. Better suited for signal processing or classification with small inputs.

Open-Awesome

MicroMLP

MITPython

A lightweight multilayer perceptron neural network library for MicroPython, designed for embedded systems like ESP32 and Pycom modules.

Visit Website GitHub

196 stars29 forks0 contributors

What is MicroMLP?

MicroMLP is a lightweight multilayer perceptron neural network library designed for MicroPython, enabling machine learning on embedded systems like ESP32 and Pycom modules. It provides a full-featured neural network implementation with support for supervised learning, reinforcement learning, and multiple activation functions, all optimized for resource-constrained environments.

Target Audience

Embedded systems developers and IoT engineers who need to implement neural networks directly on MicroPython-based microcontrollers for tasks like signal processing, control systems, or classification.

Value Proposition

Developers choose MicroMLP for its simplicity, single-file deployment, and comprehensive neural network features tailored for MicroPython, making it one of the few options for bringing ML to low-power embedded devices without heavy dependencies.

Overview

A micro neural network multilayer perceptron for MicroPython (used on ESP32 and Pycom modules)

Use Cases

Best For

Implementing neural networks on ESP32 or Pycom microcontrollers
Adding machine learning capabilities to MicroPython projects
Edge AI applications with limited memory and processing power
Educational projects teaching neural networks on embedded systems
Signal processing and classification tasks on IoT devices
Reinforcement learning experiments in resource-constrained environments

Not Ideal For

Projects requiring convolutional or recurrent neural networks for tasks like image or sequence data
High-throughput inference on servers or GPUs where performance is critical
Teams needing extensive pre-trained models or integration with popular ML frameworks like TensorFlow
Applications that rely on advanced neural network features like dropout or batch normalization

Pros & Cons

Pros

Lightweight Single-File Deployment

The library is implemented in a single `microMLP.py` file, making integration and deployment on resource-constrained microcontrollers straightforward without complex dependencies.

Comprehensive Activation Functions

Supports six activation functions including Heaviside, Sigmoid, TanH, and ReLU per layer, providing flexibility for different neural network designs as shown in the README tables.

Embedded-Focused Learning

Includes supervised learning with example management and Q-learning for reinforcement learning, specifically optimized for edge AI on devices like ESP32 and Pycom modules.

Model Persistence via JSON

Enables saving and loading entire neural network structures to/from JSON files, facilitating portability and offline use, as demonstrated in the XOR example code.

Cons

Limited to MLP Architecture

Only implements multilayer perceptrons, missing support for other neural network types like CNNs or RNNs, which restricts its use in domains requiring more complex architectures.

MicroPython-Specific Lock-In

Designed exclusively for MicroPython environments, so it cannot be used in standard Python projects, limiting its applicability to embedded systems only.

Sparse Advanced Features

Lacks modern neural network techniques such as dropout, batch normalization, or advanced optimizers, which are standard in full-scale machine learning libraries.

Frequently Asked Questions

Related Projects

emlearn-micropython

Machine Learning and Digital Signal Processing for MicroPython

Stars167

Forks20

Last commit2 months ago

upython-chat-gpt

ChatGPT for MicroPython

Stars30

Forks3

Last commit3 years ago

mp_esp_dl_models

Micropython binding for the ESP32 DL AI vision models like face detection / recognition, imagenet classifier or pedestrian (human) detection

Stars17

Forks13

Last commit1 year ago

MicroPython-NeuralNetwork

Neural Network for MicroPython

Stars0

Forks0

Last commit

Community-curated · Updated weekly · 100% open source

Found a gem we're missing?

Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.

Submit a project Star on GitHub