Contributed talk in Neural Networks 1, July 31, 2019, 11:30 a.m. in room NUBS 2.10

Evolution of metamemory ability by artificial neural networks with neuromodulation

Yusuke Yamato, Reiji Suzuki, Takaya Arita

watch Publication This study aims to evolve artificial neural networks with neuromodulation, that have a metamemory function. We show in one of the evolutionary experiments that evolved neural networks had a capacity for metamemory in the sense that they satisfy a behavioral measure based on an escape response paradigm used in animal experiments. However, it cannot exclude the solutions that change their behavior according only to particular stimuli configurations. We then define two more measures one by one responding to the study scheme (repetition of evolutionary experiments, analysis of the evolved networks and refinement of the measure). We finally obtain a neural network that successfully meets the behavioral measure, but with referring to a part of the network functioning as a memory.