Visual Cortex and Deep Networks: Learning Invariant Representations (Computational Neuroscience Series)

Visual Cortex and Deep Networks by Poggio and Anselmi is a mathematically intense explanation of a hierarchical theory of cortex processing they call "i-theory". The basis of this theory is embodied in Hubel and Wiesel's postulate of the convergence of multiple simple cells onto complex cells in a manner that generates, over some range, affine invariances, such as translation, orientation and scaling. Repetition of this transformation across multiple cortical areas in the ventral stream generates affine invariance across large areas of the receptive field. It also provides a mechanism for coupling to memory for single trial learning.Unlike many cortical theories, i-theory is well grounded in known biophysical properties of neurons. It makes specific predictions for the size of the fovea and Gabor-like cortical receptive field properties, among other factors. This is an important book that should guide cortical neuroscientists and theoreticians for many decades. Although the extensive mathematical proofs in the book are complex, the book is quite readable is one just follows the lines of argument as they are espoused.Franklin R. AmthorAuthor of "Essentials of Modern Neuroscience"

Very helpful book. :)