Hybrid Spiking Neural Network - Transformer Video Classification Model

This is the code implementation of the SSN architecture of my BSc thesis in computer science available here.

HyTSSN is a Python-based project designed for simulating and analyzing hybrid temporal-spatial spiking neural networks. The project provides modular components for building, simulating, and visualizing spiking neural network architectures.

---

Features

• Modular neural network components including synapses, dendrites, and currents.
• Temporal and spatial resolution handling for simulations.
• Customizable encoding mechanisms.
• Visualization support for model architectures.

---

Project Structure

Source Code Files

• Currents.py: Defines and manages the neural currents for the model.
• Dendrite.py: Implements dendrite functionality in the network.
• Encoder.py: Handles input encoding for network simulations.
• Input.py: Manages the input data structures and preprocessing.
• Models.py: Core implementation of the spiking neural network models.
• Synapse.py: Contains logic for synapse interactions and updates.
• TimeResolution.py: Manages the time resolution of network simulations.
• main.py: The main entry point for running simulations.

Additional Files

• Model Architecture Design.png: A diagram illustrating the architecture of the model.
• README.md: Project documentation (this file).

---

Prerequisites

• Python 3.10 or higher.
• Required Python libraries:
  • Available in requirements.txt.

---

Installation

1. Clone the repository:

   git clone https://github.com/your-username/HyTSSN.git

2. Navigate to the project directory:

   cd HyTSSN

3. Install required dependencies:

   pip install -r requirements.txt

---

Usage

1. To run the main simulation:

   python main.py

2. Customize model parameters in the respective source files (main.py, Models.py, and Encoder.py).

---

Architecture Overview

The Hybrid Spiking Neural Network - Transformer Model architecture is composed of:

• Input Layer: Inputs the encoded input stimuli using the Encoder.py module.
• Hidden Layers: Simulates the middle layers of a Cortical Column by synaptic interactions (Synapse.py) and dendritic processing (Dendrite.py).
• Output Layer: Like a Cortical Column, aggregates results to provide final results.

---

Contribution

Contributions are welcome! Please fork the repository and submit a pull request with your changes.

---

License

This project is licensed under the MIT License.

---

Contact

For questions or support, reach out to:

• Author: Aaron Bateni
• Email: [email protected]