2018/03/09 by Zhinus Marzi, Marzi, Zhinus, João P. Hespanha +3
Computer Science · Engineering · Neuroscience · #Advanced Memory and Neural Computing #FOS: Biological sciences #FOS: Computer and information sciences #Machine Learning (stat.ML) #Neural Networks and Applications #Neural and Evolutionary Computing (cs.NE) #Neural dynamics and brain function #Neurons and Cognition (q-bio.NC)
paper · pdf · doi:10.48550/arxiv.1803.03692
openalex publication_date 2018/03/09 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
There is growing evidence regarding the importance of spike timing in neural\ninformation processing, with even a small number of spikes carrying\ninformation, but computational models lag significantly behind those for rate\ncoding. Experimental evidence on neuronal behavior is consistent with the\ndynamical and state dependent behavior provided by recurrent connections. This\nmotivates the minimalistic abstraction investigated in this paper, aimed at\nproviding insight into information encoding in spike timing via recurrent\nconnections. We employ information-theoretic techniques for a simple reservoir\nmodel which encodes input spatiotemporal patterns into a sparse neural code,\ntranslating the polychronous groups introduced by Izhikevich into codewords on\nwhich we can perform standard vector operations. We show that the distance\nproperties of the code are similar to those for (optimal) random codes. In\nparticular, the code meets benchmarks associated with both linear\nclassification and capacity, with the latter scaling exponentially with\nreservoir size.\n