UNIVERSITY OF HERTFORDSHIRE
             COMPUTER SCIENCE RESEARCH COLLOQUIUM

                            presents


         "Animat Control by Spiking Neural Networks Evolved
                     with a Genetic Algorithm"

                        Prof. Borys Wrobel
           (Adam Mickiewicz University, Poznan, Poland)

                     23 April 2015 (*Thursday*)
                          11 am -12 noon

                  Hatfield, College Lane Campus
                       Seminar Room F326

                  Everyone is Welcome to Attend

                  Refreshments will be available





Abstract:

Directional movement is an example of a minimally cognitive
behaviour---a behaviour shown by even the simplest animals, and that
can be explored using simple robots. Even very simple networks (such
as natural and artificial genetic or neural networks) allow for
control of the simplest search/avoidance behaviours. Moreover, this
cognitive task can be made more difficult, so it can be seen as a
possible stepping step toward advanced cognitive skills, both in
biology and robotics.

In biology, the topology and weights in networks controlling the
behaviour of simple animals are  rather evolved than learned. I will
describe work done with an artificial life platform called GReaNs,
developed in-house. The platform implements a genetic algorithm to
evolve simple networks---artificial genetic networks or spiking neural
networks. The way the way the topology and weighs are encoded in
artificial genomes is inspired by genetic networks. This results in a
mixed bio-inspired paradigm for neural networks (the network is
encoded as if it were genetic, but works as neural). GReaNs allows
evolving animats with multicellular soft bodies (each cell with a
network with the same topology/weights, but different state), but we
have thus far only explored control of such bodies with genetic
networks. Control with spiking neural networks---on which I will
concentrate during my presentation---was explored with much simpler
animats, with pre-specified rigid circular shape. Each animat is
controlled by one network, connected to two sensors (on the left and
right front of the robot), and two actuators (which generate thrust,
on the left and right back).  Computational units in the network are
modelled as either leaky integrate and fire neurons with a fixed
threshold or adaptive-exponential integrate and fire neurons, with no
pre-specified number of neurons or synapses. I will show results of
the experiments on the evolution of such networks to control
directional movement in 2-dimensional artificial environment with
point light/sound/smell sources, and discuss the road toward more
difficult tasks, which combine navigation of the environment with
temporal/spatial pattern recognition.

---------------------------------------------------
Hertfordshire Computer Science Research Colloquium
http://cs-colloq.stca.herts.ac.uk