BOOK REVIEW David Marr’s Vision: floreat computational neuroscience VISION: A COMPUTATIONAL INVESTIGATION INTO THE HUMAN REPRESENTATION AND PROCESSING OF VISUAL INFORMATION

This is a reprint of David Marr’s 1982 book. A foreword placing the book in its historical context is added by Shimon Ullman, and an afterword by Tomaso Poggio is added on some of the themes in the book. David Marr was one of the originators of computational neuroscience, and the useful re-publication of this book enables us to assess how this field is developing, and to put David Marr’s contributions into perspective. David Marr (1945–80) obtained a First Class degree in Mathematics at the University of Cambridge in 1966; and was sufficiently interested in how the brain works to attend the Part II undergraduate courses in physiology and psychology of the Natural Sciences Tripos. (David was not experienced in practical classes, and happened to be paired with Barbara Rolls, the first female PhD student in physiology at Cambridge, who also sat in on the practical classes and provided expertise partly as a result of her training with Alan Epstein at the University of Pennsylvania.) One of the lecturers in physiology was Giles Brindley, who was interested in vision (as were many of the other members of the Department, including Horace Barlow, Fergus Campbell, William Rushton and John Robson) and in synaptic physiology. [Giles Brindley’s Physiology of the Retina and Visual Pathway (Physiological Society Monograph No. 6, Edwin Arnold, London) appeared in 1970.] Giles Brindley published a paper on how different classes of synapses might show plasticity and contribute to learning in neural networks (Brindley, 1969). These lectures and this work stimulated David’s thinking about synaptic modification and its role in systems in the brain that learn. This led to three seminal papers: on the hippocampus (Marr, 1971), the cerebellum (Marr, 1969) and the neocortex (Marr, 1970). David’s theory of the hippocampus was influenced too at the systems level by Larry Weiskrantz’s Part II lectures in psychology, which treated topics such as memory and emotion (Weiskrantz, 1956, 1968a, b; Weiskrantz and Saunders, 1984). The same Part II lectures also influenced my own research on memory, emotion and vision (Rolls, 2005, 2008). One important property of David Marr’s approach at this time was the move to take into account the quantitative network architecture of the brain system being modelled—the hippocampus, cerebellum and neocortex (Marr, 1969, 1970, 1971)—to produce a quantitative theory. This has proven to be very important in subsequent computational neuroscience approaches to memory, vision, attention and decision making (Rolls and Treves, 1998; Rolls and Deco, 2002; Rolls, 2008; Rolls and Deco, 2010). However, neuroscience was insufficiently advanced in the 1970s for David Marr to put his theories to empirical test. Nonetheless, he did try—working for example with John Eccles (Eccles et al., 1967) to test the prediction that the cerebellar parallel fibre to Purkinje cell synapses would modify associatively with the inferior olive/climbing fibre input to the Purkinje cell. They were not able to confirm this prediction, perhaps in part because the climbing fibre input was stimulated at much higher rates than these fibres are now known to fire naturally (in the range of 0–10 spikes/s). But this fundamental tenet and prediction of the theory of learning that occurs at these synapses was subsequently confirmed (Ito, 1984). Partly because of this difficulty in testing his neural network theories of cortical structure in the 1970s, David Marr chose to move to a different level of investigation in which computations being performed were studied, and tested by psychophysics, rather than being modelled at the level of their implementation in the brain. It is at this level that his 1982 book, Vision, is written. David performed the research for the book at the Massachusetts Institute of Technology (MIT) where he had moved in 1973, partly, I was told, because MIT could provide a teletype in his VISION: A COMPUTATIONAL INVESTIGATION INTO THE HUMAN REPRESENTATION AND PROCESSING OF VISUAL INFORMATION By David Marr 2010. London and Cambridge, MA: MIT Press Price: £22.95/$30 ISBN: 978-0-262-51462-0 doi:10.1093/brain/awr013 Brain 2011: 134; 913–916 | 913

[1]  M. Dawkins The Noisy Brain: Stochastic Dynamics as a Principle of Brain Function The Noisy Brain: Stochastic Dynamics as a Principle of Brain Function. By Edmund T. Rolls & Gustavo Deco. Oxford: Oxford University Press (2010). Pp. 310. Price £37.95 hardback. , 2010, Animal Behaviour.

[2]  E. Rolls,et al.  Computational models of schizophrenia and dopamine modulation in the prefrontal cortex , 2008, Nature Reviews Neuroscience.

[3]  E. Rolls Memory, Attention, and Decision-Making: A unifying computational neuroscience approach , 2007 .

[4]  Thomas Serre,et al.  Robust Object Recognition with Cortex-Like Mechanisms , 2007, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[5]  Gustavo Deco,et al.  Computational neuroscience of vision , 2002 .

[6]  Tomaso Poggio,et al.  Models of object recognition , 2000, Nature Neuroscience.

[7]  E. Rolls Functions of the Primate Temporal Lobe Cortical Visual Areas in Invariant Visual Object and Face Recognition , 2000, Neuron.

[8]  T. Poggio,et al.  Hierarchical models of object recognition in cortex , 1999, Nature Neuroscience.

[9]  E. Rolls,et al.  Neural networks and brain function , 1998 .

[10]  E T Rolls,et al.  Neurophysiological mechanisms underlying face processing within and beyond the temporal cortical visual areas. , 1992, Philosophical transactions of the Royal Society of London. Series B, Biological sciences.

[11]  L. Weiskrantz,et al.  Impairments of visual object transforms in monkeys. , 1984, Brain : a journal of neurology.

[12]  伊藤 正男 The cerebellum and neural control , 1984 .

[13]  D. Marr,et al.  Representation and recognition of the spatial organization of three-dimensional shapes , 1978, Proceedings of the Royal Society of London. Series B. Biological Sciences.

[14]  D Marr,et al.  Simple memory: a theory for archicortex. , 1971, Philosophical transactions of the Royal Society of London. Series B, Biological sciences.

[15]  D. Marr A theory for cerebral neocortex , 1970, Proceedings of the Royal Society of London. Series B. Biological Sciences.

[16]  G S Brindley,et al.  Nerve net models of plausible size that perform many simple learning tasks , 1969, Proceedings of the Royal Society of London. Series B. Biological Sciences.

[17]  D. Marr A theory of cerebellar cortex , 1969, The Journal of physiology.

[18]  L. Weiskrantz Experiments on the r.n.s. (real nervous system) and monkey memory , 1968, Proceedings of the Royal Society of London. Series B. Biological Sciences.

[19]  L Weiskrantz,et al.  Analysis of Behavioural Change , 1968 .

[20]  Professor Dr. John C. Eccles,et al.  The Cerebellum as a Neuronal Machine , 1967, Springer Berlin Heidelberg.

[21]  L. Weiskrantz,et al.  Behavioral changes associated with ablation of the amygdaloid complex in monkeys. , 1956, Journal of comparative and physiological psychology.