Overview

This thesis focuses on implementing and evaluating spiking neural networks (SNNs) in a software-defined framework. The goal is to develop computational models of analog spiking neurons based on published behavior (e.g., integrate-and-fire dynamics, threshold adaptation, refractory periods) and utilize these models to simulate neuromorphic networks for tasks such as pattern recognition, filtering, or classification.
The work involves building simulation tools or leveraging existing frameworks (e.g., SNNTorch, Brian2, Nengo) to define spiking neurons as programmable units whose firing patterns depend on input events and tunable parameters (e.g., spiking thresholds, input weights, delays). These neurons will then be combined into functional networks, demonstrating how spiking activity can encode and process information without traditional machine learning components.

Key Features

      • Develop software models for spiking neurons with tunable dynamics (threshold, refractory time, etc.).
      • Build and simulate functional SNNs using those neuron models.
      • Implement computational tasks (e.g., binary classification, temporal filtering) using spike-based coding.
      • Visualize network behavior (e.g., spike behavior, frequency responses, decision outcomes).
      • Analyze performance in terms of accuracy, responsiveness, and parameter sensitivity.

Methods

      • Use Python and frameworks like Brian2, Nengo, or custom code.
      • Encode inputs as spike trains (e.g., rate or temporal encoding).
      • Model spike generation and event-driven network dynamics.
      • Evaluate different network topologies and parameter settings for specific tasks.
      • Present results through visual analytics (plots, animations, confusion matrices).

Skills acquired within the thesis

     • Foundations of neuromorphic computing and spike-based computation.
     • Practical experience with SNN simulation tools and frameworks.
     • Software modeling of time-based neural dynamics.
     • Understanding of computational neuroscience principles in a programmable context.
     • How to build and test bio-inspired algorithms using event-driven logic.

Prerequisites

     • Strong programming skills in Python
     • Familiarity with basic neural network or machine learning concepts
     • Interest in event-driven systems, computational models, or biologically inspired AI

References:

1. https://brian2.readthedocs.io/en/stable/
2. https://snntorch.readthedocs.io/en/latest/
3. https://www.nengo.ai/
4. Flexible Unipolar IGZO Transistor-Based Integrate
5. Analog Printed Spiking Neuromorphic Circuit