This project focuses on developing a Bio-Inspired CODEC using a custom Spiking Neural Network (SNN) to process image and sound data from videos. The SNN is designed with multi-layered encoding channels to compress and recover signals efficiently. By introducing noise in intermediate layers, the network is tested under realistic conditions, optimizing its performance using synaptic learning and spike metrics for error quantification.
- Multi-Layered SNN: Custom SNN architecture with encoding channels for image and sound data.
- Noise Simulation: Real-world conditions simulated with noise introduction in intermediate layers.
- Synaptic Learning: Decoding layer engineered with synaptic learning for efficient signal recovery.
- Spike Metrics: Advanced spike metrics used for precise error quantification.
- NEST for SNN simulation
- Python for implementation and data processing
- Advanced Neural Network Architectures
- Signal Encoding and Decoding
- Synaptic Learning Techniques
- Temporal Encoding Methods
- AI Ethics and Analytical Skills
Image and audio data extraction from videos. Encoding signal and noise simulation.
Custom SNN architecture design with noise handling. Training and testing using separate datasets for robust evaluation.
Performance metrics including spike metrics. Ethical considerations and AI bias mitigation.
Clone the repository:
- Launch ipynb file that contains already run results
- bash
- Copy code files
- pip install -r requirements.txt
- Configure SNN parameters using JSON files in the config directory.
- Run the main notebook to start the simulation:
- bash
- copy code
- run Spiking Neural Network - Bio-Inspired CODEC.ipynb
- Enhancing SNN architecture for better performance in diverse conditions.
- Expanding the dataset for improved generalization.
- Further integration of ethical AI practices and bias mitigation.