Friday, 17th of January

14h30 (room R2014, 660 building) (see location)

Note: date changed to Friday 17th

Loris Felardos

(TAU)

Loris presents the paper:

Biologically inspired alternatives to backpropagation through time for learning in recurrent neural nets

I'll present this paper: https://arxiv.org/abs/1901.09049
which has received overwhelmingly positive reviews at NeurIPS 2019 [1].
It solves the fundamental problem of efficient online learning (no BPTT)
in the context of Spiking Neural Networks [2] and was presented during
the Neuro-AI workshop [3]. I'll introduce all the prerequisites so you
don't need to know anything about Spiking Neural Networks.

Refs

[1] Reviews are here: https://openreview.net/forum?id=SkxJ4QKIIS
[2] The problem has been solved for artificial (non-spiking) neural networks in another paper which I'll discuss as well.
[3] Workshop link: https://neurips.cc/Conferences/2019/Schedule?showEvent=13189


All TAU seminars: here