Can Mekik, RPI Graduate Student

 

Can Mekik, RPI Graduate Student

Sage 4101

September 7, 2016 12:00 PM - 1:30 PM

Imagination tasks are tasks that, if performed by humans, would require the use of imagination. Artificial imagination systems are systems which perform imagination tasks. These systems are applicable in a variety of settings such as procedural generation, interactive search and robotics (among others), as well as in cognitive modeling.

Evaluation is a pervasive problem in imagination systems research. Many existing evaluation procedures are ad hoc in that they are not motivated by a general theory of imagination. Ad hoc evaluation procedures make it difficult to interpret results and to compare systems. In some settings, it is possible to recruit human assessors for imagination system evaluation. The use of human assessors often allows researchers to bypass the difficulties associated with ad hoc evaluation procedures, but running experiments with human assessors is resource intensive.

It may be possible to develop improved evaluation procedures for imagination systems by implementing artificial assessors. In this talk, I present a framework for artificial assessor development which integrates insights from the philosophy and psychology of imagination as well as from imagination systems research. I then report the results of a pilot artificial assessor development project and I argue that artificial assessor implementation is a tractable and scientifically rewarding enterprise.

 

Add to calendar
Share|