Brain-Like Evolving Spiking Neural Networks for Multimodal Information Processing

Abstract

Despite of much evidence suggesting how and where sensory information converge in the human brain, the neural mechanisms of interaction among modalities at the level of neuronal cells and ensembles are still not well understood. The chapter explores emulation of multimodal information processing in a brain-like manner through evolving spiking neural network (ESNN) architectures that use several multimodal characteristics of the biological brains, e.g., multisensory neurons, crossmodal connections, capacity of lifelong adaptation and evolution, adaptive pattern recognition. Illustration is given on audiovisual ESNN for the person authentication problem. Preliminary results show that the integrated system can improve the accuracy in many operation points as well as it enables a range of multi-criteria optimizations.