Zoran Tiganj¹, Karthik Shankar², Marc Howard³
¹Boston University, ²Boston University, ³Boston University
Natural signals, including auditory, visual and natural language signals, show autocorrelations over a wide range of time scales. As a response to this rich structure, different regions of the human brain contain temporal receptive windows that respond to features across a range of distinct temporal scales, up to tens of seconds. Here we propose a neutrally plausible computational model that provides a frequency decomposition of the input signal and that can account for such temporal windows. The model continues on a previously proposed model that accounts for the maintenance of a scale-invariant representation of the input history. We show that by combining a set of cells that operate as low-pass filters we can obtain a set of cells that operate as band-pass filters with some interesting properties. The width of the pass-bands increases progressively with the central frequency of the filters in a scale-invariant fashion. We show that these band-pass filters constitute the temporal receptive windows resembling those observed in early sensory areas as well as those observed in higher cortical regions.
Keywords: Receptive Windows, Slowness