April 24, 2014 14:00 〜 15:30
CiNet 1F Conference Room A
Bio-inspired computing systems are the candidates of next-generation energy-efficient information systems.
Many efforts have been devoted to the development of brain-like computing systems with low power consumption.
However, there is still a large gap between the well-designed architecture of biological neuronal networks and the idealized structure of artifical neural networks. We are trying to develop artifical neural networks which can be implemented using devices with less energy consumption. In general, when we reduce the number of neurons and/or interconnections between neurons, the information processing performance of neural networks will decrease. Therefore, the problem is related to the tolerance of computing ability of the network against component failure. In this talk, I focus on the concept of network robustness and its application to a development of energy-efficient neural networks. In the first part, I introduce the theoretical studies on dynamical robustness of complex networks, by which we can understand the robustness and fragility of dynamical behavior in populations of interacting dynamical units. In the second part, I talk about recent challenges for realizing energy-efficient artifical neural networks.