AER Spiking Neuron Computation on GPUs: The Frame-to-AER Generation

Abstract

Neuro-inspired processing tries to imitate the nervous system and may resolve complex problems, such as visual recognition. The spike-based philosophy based on the Address-Event-Representation (AER) is a neuromorphic interchip communication protocol that allows for massive connectivity between neurons. Some of the AER-based systems can achieve very high performances in real-time applications. This philosophy is very different from standard image processing, which considers the visual information as a succession of frames. These frames need to be processed in order to extract a result. This usually requires very expensive operations and high computing resource consumption. Due to its relative youth, nowadays AER systems are short of cost-effective tools like emulators, simulators, testers, debuggers, etc. In this paper the first results of a CUDA-based tool focused on the functional processing of AER spikes is presented, with the aim of helping in the design and testing of filters and buses management of these systems.