Читать книгу The Behavior of Animals - Группа авторов - Страница 54

Building on the neuroethological results of the toad’s visual system (e.g., Figure 2.10), artificial pattern recognizers were developed—using systems theoretical approaches (Ewert & v. Seelen 1974; cit. Ewert 1984)—computer models taking advantage of the relevant cytological brain structures (Lara et al. 1982), and artificial neuronal nets, ANNs, trained by backpropagation algorithms (Ewert 2004). Different ANNs applying algorithms for reinforcement learning, classical contitioning, and genetic operations are described by Reddipogu et al. (2002) and Yoshida (2016). Hence, there are various ways of modeling brain/behavior functions: global models are heuristic; ANNs subserve approximation and optimization, e.g., by implementation of an algorithm.

Why modeling? 1) A model offers a representation of the processes within the modeled system. Hence, models have explanatory function. 2) Models are predictive. Predictions can be tested by adequate experiments. The results, in turn, may improve the model. 3) Models are sort of creative since they may exhibit unexpected properties. 4) Models provide tools toward artificial intelligence, such as in the growing field of neuroengineering.

For example, the German Federal Ministry for Research and Technology (BMFT) supported a joint project called “Sensori-Motor Coordination of Robotic Movements with Neuronal Nets” SEKON established in 1991–1994 by scholars from neurobiology, neuroinformatics, and robotics. To study interfaces between perception and action, in one experimental platform a modular structured ANN simulating toad vision (Fingerling et al. 1993)—in connection with a CCD-camera—instructed a robot to select and pick out differently shaped work pieces from a conveyor belt (see also Further Reading, Movie A1).