BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Robotics Seminar Series
SUMMARY:Neuroevolutionary approaches to generating increas
 ingly intelligent behaviours in virtual creatures 
 / (simulated) robots - Alastair Channon\, Universi
 ty of Keele
DTSTART;TZID=Europe/London:20191206T140000
DTEND;TZID=Europe/London:20191206T150000
UID:TALK135568AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/135568
DESCRIPTION:Dr Channon and his Evolutionary Systems group carr
 y out research into the use of evolution to genera
 te increasingly intelligent agents. Their work has
  included evolving deep neurocontrollers (deep neu
 roevolution) since 1996\, using neural development
  (augmenting topologies) and generative encodings 
 within that and other work also since 1996\, and i
 ncorporating (initially static\, hand-designed and
  later evolvable) convolutional neural networks wi
 th rectified linear units (ReLUs) since 2005. This
  talk will give an overview of some incremental ap
 proaches to the evolution of neural networks for t
 he control of intelligent agents / virtual creatur
 es / simulated robots. The focus here will be on '
 long-term' evolution and approaches that generate 
 increasingly intelligent behaviours. This part of 
 the talk will conclude with a very brief overview 
 of related work on open-ended evolution\, in which
  the aim is to achieve ongoing adaptive novelty an
 d ongoing growth of complexity.\n\nThe second part
  of the talk will focus on two specific examples o
 f the incremental neuroevolution of virtual creatu
 res / simulated robots.  In the first example\, a 
 simple simulated quadruped is evolved to walk over
  obstacles (walls) of different heights. A range o
 f different evolutionary complexification strategi
 es are compared and 'heterogeneous strategies' are
  shown to overcome previous approaches' shortcomin
 gs in relation to loss-of-gradient and over-fittin
 g (analogous to catastrophic forgetting in neural 
 systems).  The second example demonstrates the inc
 remental neuroevolution of reactive and deliberati
 ve 3D Agents.  The 3D physically-based setting req
 uires that a successful agent continually and deli
 berately adjust its gait\, turning and other motor
  control over the many stages and sub-stages of th
 ese tasks\, within its individual evaluation.  Ach
 ieving such complex interplay between motor contro
 l and deliberative control\, within a neuroevoluti
 onary framework\, is the focus of this work. The r
 esults demonstrate a variety of intricate lifelike
  behaviours being used\, separately and in combina
 tion.
LOCATION:James Dyson Building\, Teaching Room\, main Engine
 ering site
CONTACT:Amanda Prorok
END:VEVENT
END:VCALENDAR