Contrast invariant feature selectivity in balanced random networks

Abstract

Event Abstract Back to Event Contrast invariant feature selectivity in balanced random networks Sadra Sadeh1*, Stefano Cardanobile1 and Stefan Rotter1 1 Bernstein Center Freiburg & Faculty of Biology, University of Freiburg, Germany Neurons in the primary visual cortex (V1) of mammals are highly selective for the orientation of a light bar [1, 2, 3], while they receive afferent connections from neurons in lateral geniculate nucleus (LGN) which are themselves not selective. The emergence of this property in cortex has been subject of debate since decades, and served as a paradigm to unravel the key mechanisms of information processing in the brain. We investigate the problem from a network point of view, by focusing on random recurrent networks with no specific pattern of connectivity. By exploring a large-scale network model of spiking neurons through simulations and theoretical analysis, we show how a purely random network operating in the inhibition dominated regime contributes to feature selectivity. In particular, we demonstrate how tuning amplification can happen in these networks as a consequence of 'selective attenuation', a general mechanism which selectively suppresses the common mode. By systematically investigating the relevant parameter space, we pinpoint different regimes of orientation selectivity, which yields testable predictions for the biological cortex. The 'selective attenuation' mechanism also yields contrast invariance [4, 3] of feature selectivity, which ensures feature detection for a wide range of input scales. Therefore, we suggest that the basic mechanism of contrast invariance is a consequence of intracortical interactions, and neither a single cell property, nor a purely feedforward mechanism. We argue in favor of 'tuning amplification' as the key process of the recurrent processing stage, for which no specific structure is needed. This mechanism could, in principle, also work for other sensory features and other sensory modalities as well, since the neural hardware necessary to achieve it is readily available in the cortex. Acknowledgements Funding by the German Ministry of Education and Research (BCCN Freiburg, grant 01GQ0420 and BFNT Freiburg*Tubingen, grant 01GQ0830) is acknowledged.