SNN Conversion Demonstrator

Report Bug · Request Feature

Green-AI Hub Mittelstand

Homepage | Contact

Overview

This project provides an interactive dashboard for loading neural models, running inference on both SNN and ANN architectures, monitoring energy consumption, and streaming simulated sensor data. It's designed to showcase the benefits and trade-offs between traditional neural networks and neuromorphic computing approaches.

Features

• Neural Model Comparison: Run and compare inference between SNN and ANN models
• Real-time Energy Monitoring: Track energy consumption during model execution
• Sensor Data Simulation: Stream simulated sensor readings for testing
• Interactive Web Dashboard: Control experiments through an easy-to-use interface
• Parallel Execution: Run ANN and SNN inference simultaneously for direct comparison

Installation

Prerequisites

• Python 3.7+
• Flask
• Samna (neuromorphic computing library)
• Access to Xylo IMU hardware for energy monitoring

Setup

1. Clone this repository:

   git clone https://github.com/yourusername/SNN-Conversion-Demonstrator.git
   cd SNN-Conversion-Demonstrator
   

2. Install dependencies:

   pip install -r requirements.txt
   

3. Ensure you have the necessary models in the models/ directory.

Usage

1. Start the application:

   python app.py
   

2. Access the web interface at http://localhost:5004

3. Using the interface:

   • Load a model with your chosen parameters
   • Run inference to compare SNN vs ANN performance
   • Monitor energy consumption during execution
   • View simulated sensor data

Architecture

The application is structured around several key components:

• NeuralMonitorApp: Main Flask application that coordinates all components
• NeuralModel: Handles loading and running both SNN and ANN models
• EnergyMonitor: Tracks power consumption on neuromorphic hardware
• SensorMonitor: Provides simulated sensor data for testing

API Endpoints

• GET /: Main dashboard interface
• GET /energy: Get current energy readings
• GET /sensor_data: Get next sensor reading
• POST /load_model: Load a neural model with specified parameters
• POST /start_inference: Run inference on both ANN and SNN models
• POST /start_baseline: Trigger baseline energy measurement
• GET /get_baseline: Retrieve stored baseline energy value

Model Support

The application supports different neural network models:

• Training Mode: Uses models/SynNet_kaggle_{timesteps}timesteps_20epochs.json
• Inference Mode: Uses models/converted_rp_model_20epochs_{timesteps}timesteps_kaggle.json

Hardware Requirements

For full functionality, you'll need access to:

• Xylo IMU TestBoard for accurate energy measurements
• Sufficient compute resources for neural network inference

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT License