A Model of Amoeba-Based Neurocomputer

概要

論文の詳細を見る
An amoeboid organism, the true slime mold <span class="x-2">Physarum</span>, has been studied actively in recent years to explore its spatiotemporal oscillatory dynamics and various computational capabilities. In the authors previous studies, the amoeba was employed as a computing substrate to construct a neurocomputer. Under optical feedback control to implement a recurrent neural network model, the amoeba grows or withdraws its photosensitive branches by exhibiting a number of spatiotemporal oscillation modes in search of a solution to some combinatorial optimization problems. In this paper, considering the amoeba as a network of oscillators that compete for constant amounts of resources, we model the amoeba-based neurocomputer. The model generates several oscillation modes and produces not only simple behavior to stabilize a single mode but also complex behavior to spontaneously switch among different modes. To explore significances of the oscillatory dynamics in producing the computational capabilities, we establish a test problem that is a kind of optimization problem of how to allocate a limited amount of resource to oscillators so that conflicts among theoscillators can be avoided. We compare the performances of the oscillation modes in solving the problem in a bottom-up manner.