Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Brain computation as hierarchical abstraction
Ballard D., The MIT Press, Cambridge, MA, 2015. 456 pp.  Type: Book (978-0-262028-61-5)
Date Reviewed: Sep 15 2015

The Turing Award winner and Nobel laureate Herbert A. Simon stated, in his landmark work on complex systems [1], that hierarchy “is one of the central structural schemes that the architect of complexity uses.” Frequently, in both natural and artificial systems, “complexity takes the form of hierarchy--the complex system being composed of subsystems that in turn have their own subsystems, and so on” [1]. Simon argued that hierarchic systems evolve far more quickly than non-hierarchic systems of comparable size, and illustrated this fact by using a parable of two watchmakers. Both made watches of 1000 pieces; one used subassemblies of ten elements that could be put together into larger subassemblies, and the other’s watches fell to pieces and had to be reassembled from scratch as soon as there were any interruptions in the assembly process. It is not surprising that the first one prospered in the parable while the second one filed for bankruptcy. Likewise, trying to understand how the human brain works using hierarchical descriptions of the computational processes it performs is also unsurprising.

Dana Ballard draws a parallelism between the different abstraction levels in computer systems, from the logical gates in its hardware to its software applications, and a tentative decomposition of neural computation into different levels of abstraction. Obviously, brains are completely different than silicon computers and must perform “tricks” (sic) to overcome their extremely slow speed when compared to their artificial counterparts, letting apart any discussion of the potential non-Turing abilities of humans. Once the stage is set, Ballard devotes a second introductory chapter to summarize what is known about the major subsystems of the brain and their interactions, with a special emphasis on the mammalian forebrain.

After the introductory chapters, the next part of the book focuses on the bottom of the neural computational hierarchy: neurons and their circuits within the brain. The analysis of the slow brain signaling mechanisms, that is, spikes, paves the way for a discussion of how learning can explain the astonishing plasticity of the mammalian cortex. A 14-page appendix also describes how individual neurons work: the action potential, the anatomy of synapses, and Hebb’s learning rule. An individual chapter is devoted to the lookup-table organization of the cortex, which acts as a content-addressable memory. Finally, the closing chapter delves into the execution of neural programs, described in terms of states and actions, whose execution seems to be controlled by the basal ganglia in the brain. The search for the right program to execute, and its scoring or evaluation, is explained in terms of simple reinforcement learning (Q-learning or temporal-difference learning for those acquainted with artificial intelligence (AI) terminology), whereas the learning needed to set cortical synapses is explained in terms of the Widrow-Hoff learning rule.

The previous chapters focused on the brain circuits by putting aside the brain’s interaction with the outside world, which is exactly what you might expect given the book’s title. However, the second half of the book turns its attention to its interface with the rest of the world. First, its embodiment is analyzed with respect to its sensors (especially vision) and its actuators (muscles). Visual routines exploit context to interpret the noisy input the brain receives from the specialized human vision systems, given its anatomical peculiarities and its different gaze stabilization mechanisms. Motor routines help us solve the reverse problem, by controlling the motor system with the help of the cerebellum. The visual and motor routines abstraction helps explain complex behavior, which is cast as a concurrent and dynamic set of indexable routines, which are amenable again to reinforcement learning in the brain.

The final chapters raise the level of abstraction and analyze brain computation in a social context. When other agents are present, the author resorts to game theory in order to explain social decision making. The brain’s emotional system and the issue of consciousness also get their share of attention in the final two chapters, which are necessarily more tentative and vague in their conclusions.

In summary, this book succinctly surveys what is known about the human brain from the perspective of hierarchies by combining insights from many different researchers and a wide range of experimental results that provide psychophysical evidence for particular theories. While many discussions are unavoidably shallow, most stances are open to debate and many assertions will be proven wrong in the future. However, even though many issues are far from settled from a scientific point of view, the author draws suggestive and insightful analogies (such as that of the neuro, the neural currency used as a reward mechanism in reinforcement learning, which is associated with the dopamine neurotransmitter). The book is extremely well written, includes excellent illustrations, and is easy to follow for non-neuroscientists, especially if you take into account that delving into the complex details of brain mechanisms tends to blur the big picture (and makes neuroscience books hard to swallow).

Reviewer:  Fernando Berzal Review #: CR143774 (1512-1031)
1) Simon, H. A. The sciences of the artificial (3rd ed.). MIT Press, Cambridge, MA, 1996.
Bookmark and Share
  Reviewer Selected
Featured Reviewer
 
 
Biology And Genetics (J.3 ... )
 
 
Cognitive Simulation (I.2.0 ... )
 
Would you recommend this review?
yes
no
Other reviews under "Biology And Genetics": Date
Computational phylogenetics: an introduction to designing methods for phylogeny estimation
Warnow T.,  Cambridge University Press, New York, NY, 2018. 394 pp. Type: Book (978-1-107184-71-8)
Nov 14 2018
Computing the expected value and variance of geometric measures
Tsirogiannis C., Staals F., Pellissier V.  Journal of Experimental Algorithmics 23(2): 1-32, 2018. Type: Article
Nov 9 2018
Mathematical grammar of biology
Yamagishi M.,  Springer Publishing Company, Incorporated, New York, NY, 2017. 82 pp. Type: Book (978-3-319626-88-8), Reviews: (2 of 2)
Jul 13 2018
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright © 2000-2018 ThinkLoud, Inc.
Terms of Use
| Privacy Policy