Sabre Didi awarded runner-up for best paper at SSCI 2016 doctoral consortium

Best Student Paper Runner up Award at IEEE Symposium Series on Computational Intelligence (SSCI) 2016 held in Athens, Greece. SSCI is a flagship annual international conference on computational intelligence sponsored by the IEEE Computational Intelligence Society, promoting all aspects of theory, algorithm design, applications and related emerging techniques (http://ssci2016.cs.surrey.ac.uk/). The SSCI conference hosted about 350 paper presentations, of which 30 accepted papers, first authored by PhD students were selected to be presented at the Doctoral Consortium.

Sabre Didi with his PhD supervisor Dr. Geoff Nitschke, investigated the impact of coupling transfer learning with behaviour diversity maintenance methods on controller evolution in multi-robot tasks. Multi-robot task control is a challenging problem for Evolutionary Algorithms (EA) using a single objective function. A significant body of research has shown that EA suffer from premature convergence and fails to find optimum solutions during the evolutionary synthesis of multi-robot controllers. EA policy search suffers from two main issues inherent to learning in multi-robotics: the bootstrap problem and deception. The bootstrap problem occurs when a task is too demanding for an objective function to apply meaningful selection pressure on randomly initialised population of solutions. Deception occurring when evolution fails to build a gradient that leads to a global optimum and instead is driven towards poor solutions. As a result EA converges prematurely to suboptimal solutions. The paper highlights the superiority of coupling transfer learning with a hybrid of behaviour diversity maintenance and objective based search in avoiding the bootstrap problem and deception in multi-robot tasks.