Developing Intelligence

Is there a basic “computational unit” of the neocortex? In contrast to subcortical regions, neocortical architecture seems fairly regular and matrix-like – leading to it’s other name, “isocortex.” While there are many contenders for the title of the “canonical circuit” or “cortical algorithm”, few would dispute that cortical columns are a fundamental organizational principle of cortex.

Or wouldn’t they? In their new Neuron article, Douglas & Martin’s argue that the cortical column is a poor contender for identifying the cortex’s “canonical circuit.” They describe how the anatomical boundaries between functional columns are not clear: although cell bodies line up in orderly columns, the connectivity amongst these cell bodies does not seem to respect any boundary. At the same time, electrophysiological investigations are quite consistent in showing a columnar functional organization. In their words, “there is simply a mismatch between the anatomy and the functional maps.”

In Douglas & Martin’s view, there are better candidate principles for the neuroanatomical organization of neocortex: they review evidence for a front-to-back gradient in the size and dendritic reach of pyramidal cells among old world monkeys, a circuit of lateral connectivity in cortex called a “daisy” (present in all species except rodents), an apparently universal correlation across species between the size and spacing of these “daisy” structures, and a conserved ratio of spiny excitatory neurons, and the number of synapses formed by them, relative to smooth inhibitory neurons across species and Brodmann areas. Interestingly, Brodmann area 17 (primary visual cortex) seems to violate other candidate principles, such as the ratio of connectivity to neuropil size.

Rather than getting stuck in these anatomical peculiars, Douglas & Martin argue that it’s more productive to understand the “neocortical algorithm” at a more abstract, information processing level. Effectively, they propose three anatomical principles for neocortex:

1) Just enough – it seems that thalamocortical connectivity is tuned to provide “just enough” excitatory activity to a column, in relatively small numbers of afferent connections, and that intracortical inhibition may play a relatively minor role in the functional organization of cortical columns.

2) Just in time – Douglas & Martin argue that the brain is an asynchronous computer, with no global clock to synchronize computations. Instead, few neurons are active at any given time (suggestive of “on-demand” event-based processing). The activity of these neurons intrinsically reflects time, in that it is dependent on coactivation excitatory inputs within a 10-20ms time window.

3) Recurrent connectivity – although not explicitly proposed as a principle, it’s clear that recurrent connectivity plays a large role in Douglas & Martin’s scheme. The idea is that “just enough” thalamocortical connectivity provided “just in time” can bias ensembles of neocortical neurons, to accomplish pattern completion and constraint satisfaction.

Douglas & Martin suggest that these principles make up the “syntax” of cortex, which varies in its particulars as much as the world’s languages. Only by examining these grander structural codes can we correctly translating the brain’s functional characteristics into a detailed understanding of anatomical circuitry.


  1. #1 Eric Thomson
    December 17, 2007

    I wrote up a follow-up to this interesting post at my blog. Thanks for the pointer to this paper: it really drives home how much we have yet to learn.

The site is currently under maintenance and will be back shortly. New comments have been disabled during this time, please check back soon.