THOR Hands-On Tutorial

Neuromorphic Computing at Scale: A practical, step-by-step guide to accessing and utilizing the large-scale neuromorphic infrastructure.

Objectives

  • Spike-based Processing: Learning about computing with spikes and spiking neural networks.
  • Ecosystem Navigation: Understand the THOR hardware-software landscape.
  • Simulation and Modeling: Configuring neuron models and running simulations.
  • Model Deployment: Execute a spiking neural network on THOR-available platforms.

Prerequisites

  • Programming: Proficiency in Python.
  • Machine Learning: Basic understanding of Neural Networks (SNN knowledge is a plus).

Hardware Platforms Covered

SpiNNaker 2

A massively parallel digital neuromorphic system featuring 152 ARM Cortex-M4F cores per chip with dedicated ML accelerators. THOR hosts 54 SpiNNaker2 server boards, each equipped with 48 SpiNNaker2 chips. Thus, giving you access to nearly 400,000 highly parallel, event-based processing elements.

Getting Started

1

Access Credentials

Collect your login credentials (IP_address, Password) from the organizers.

IP_address & Password
2

Hardware Assignment

Collect your assigned port number (PPPP).

PPPP
3

Network Connection

Connect to the Event/Guest Wifi network.

4

Setup Port Link

Run the following command in your terminal:

ssh -L PPPP:localhost:PPPP thor@IP_address

Enter the assigned Password when prompted.

IMG : Keep this terminal open in the background to keep the connection alive!

5

Access THOR Tutorial

Open your browser and navigate to:

http://localhost:PPPP

Enter the same Password to login to Jupyter Lab

Tutorial Modules

GC

THOR SNN Tutorial

Introduction to the fundamentals of spike-based information processing and SNNs.
00

SpiNNaker2 Hardware

Learning about the SpiNNaker2 Neuromorphic Platform and its software ecosystem.
01

Poisson Simulation

Implementing a poisson spike simulation on the SpiNNaker2 hardware platform.
02

Neuron Models : LIF

Exploring the Leaky Integrate-and-Fire (LIF) neuron and simulating its dynamics.
03

Neuron Models : Izhikevich

Exploring the Izhikevich neuron model and comparing the backends.
04

NIR Mapping

Mapping models to SpiNNaker 2 using the Neuromorphic Intermediate Representation (NIR).
05

SNN Deployment

Running pretrained models on the SpiNNaker 2 hardware platform.