Exploring unsupervised online learning in digital spiking neural networks
In this workgroup, we will explore unsupervised online learning with the ODIN 28-nm digital spiking neural network, if possible applied to real-world examples (e.g., the MNIST dataset).
Compared to analog approaches, digital implementations of online learning rules:
- do not suffer from variations in operating voltage or temperature,
- allow one-to-one hardware/software correspondence,
- keep the stochasticity and random-selection properties for Poisson-distributed input spike trains.
ODIN is an online-learning digital spiking neural network in 28nm CMOS recently developed at Université catholique de Louvain.
As only one chip testbench is available for in silico experimentation, a custom Python simulator of ODIN will be shared for quick exploration of parameters and network topologies.
The wiki page contains a preprint of ODIN together with useful papers and links.
|Wed, 25.04.2018||15:00 - 16:00||Sala Panorama (Hotel 1st floor, above main hall)|
|Thu, 26.04.2018||22:00 - 23:00||Sala Panorama|
|Fri, 27.04.2018||14:00 - 15:00||Lecture Room|
|Mon, 30.04.2018||15:00 - 16:00||Sala Panorama|