UNIVERSITY OF HERTFORDSHIRE
             COMPUTER SCIENCE RESEARCH COLLOQUIUM

                            presents


         "From Uninformed to Informed Exploration in Partially-
                 observable Time-varying Spaces"

                       Allan Max Axelrod
                  (Oklahoma State University, USA)

                     8 July 2015 (Wednesday)
                         10 am - 11 am

                  Hatfield, College Lane Campus
                        Seminar Room B154 

                  Everyone is Welcome to Attend

                  Refreshments will be available





Abstract:

A critical challenge to overcome in robotics is maintaining situational
awareness in real-world environments, particularly those containing
humans. Real-world environments are often partially-observable and
time-varying, yet very few robotic algorithms are designed to operate
in such circumstances. For this reason, many robotic applications fail
to provide meaningful performance guarantees outside of controlled or
stationary environments. Herein, a formulation inspired by photometrics
and human learning studies is used to develop the Real-time Adaptive
Prediction of Time-varying and Obscure Rewards (RAPTOR) algorithm. The
RAPTOR algorithm is shown to optimize situational awareness in the context
of the pure exploration problem for partially-observable and time-varying
real-world environments. Moreover, an exponential probabilistic bound
on the performance of RAPTOR is developed. This bound is used to
transition from RAPTOR from an initially naive exploration behavior to
an information-driven exploration behavior, as would be necessary in new
environments. This is in line with the numerous studies on human learning,
which have found that learning contains both naive and information-driven
behaviors.

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