Self-organising neural networks for computer vision
by Petrut Bogdan

Abstract: Biological neural circuits can be very adaptable and evolve
their dynamics to better deal with their embodied realities (through
the acquiring of memories, perfecting of motor actions and recovering
from medical conditions such as stroke, to name a few). One dimension
of this adaptability, also called plasticity, that I will focus on is
called structural plasticity. It is the continuous process of synaptic
rewiring resulting in the change of connectivity in a network of
neurons. During my PhD I have enabled the simulation of this mechanism
on neuromimetic hardware in the context of experiments involving
neural topographic maps, moving bars and handwritten digits. This
presentation will focus on 1) biological evidence of structural
plasticity and its implications, 2) results of (some of) the
previously mentioned experiments using spiking neurons and 3) applying
this computational mechanism in Deep Neural Networks, such as
MobileNet.