Introduction to Spiking Neural Networks, Neuromorphic Computing and Computational Neuroscience

In this talk, I give an introduction to spiking neural networks and neuromorphic computing. We have made incredible progress with deep learning, but there are still things that biological brains do remarkably well, such as learning continuously from experience, adapting to new situations, and operating with very little energy. Can we learn something from biology to build better AI systems? I start with some background on computational neuroscience and how biological neurons work, then go into spiking neural networks, different neuron models, and learning mechanisms such as STDP and eligibility traces. We also look at neuromorphic hardware, the current limitations of these approaches, and some of the open research problems. Toward the end, I show a couple of my own robotics experiments: a small wheeled robot that learns to chase or avoid a ball, and a quadruped that learns to compensate for additional weight on one of its legs. Both use local learning on small microcontrollers, without backpropagation. The talk is intended for people with some technical background, particularly in software engineering or machine learning, but no prior knowledge of neuroscience is needed. I'm still exploring this field myself, and one of the reasons I wanted to give this talk was to share what I've learned so far and hopefully get more people interested in it. If you have questions, ideas, or are working on something similar, I'd love to hear from you in the comments. You can also find me here: LinkedIn:   / markovelic   X: https://x.com/MarkoVelich GitHub: https://github.com/velicm 00:00 Introduction: Why NeuroAI? 04:10 What Is NeuroAI? 06:01 Computational Neuroscience Across Scales 07:53 Three Generations of Neural Networks 12:32 The Limitations of Deep Learning and Backpropagation 15:43 How Biological Brains Compute 19:54 Learning from Biological Intelligence 21:46 Brains as Dynamical Systems 26:11 Artificial vs. Biological Neural Networks 28:04 Spiking Basics and Event Cameras 29:37 Leaky Integrate-and-Fire (LIF) Neuron 33:56 Advanced Neuron Models 40:07 Dendrites and Complex Neural Dynamics 44:14 Encoding Data as Spikes 46:26 Training SNNs and Surrogate Gradients 49:01 STDP and Local Learning: A Robot Example 52:42 ANN-to-SNN Conversion 53:40 SNN Software Frameworks 54:55 Neuromorphic Hardware 57:12 Why Spiking Neural Networks Are Hard to Scale 59:45 Research Opportunities and Future Directions 1:03:10 Practical Applications of NeuroAI 1:04:20 Learning Resources and Communities 1:05:21 Robot Demo: Learning to Chase or Flee a Ball 1:06:46 Robot Demo: Adaptive Walking with a Leg Weight 1:08:12 Summary and Closing #NeuroAI #SpikingNeuralNetworks #neuromorphiccomputing