Gradient Expectations by Keith L. Downing

Gradient Expectations by Keith L. Downing

Author:Keith L. Downing
Language: eng
Format: epub
Publisher: The MIT Press


The basic design of these columns is preserved across the entire cortex and across species as diverse as mice and humans. Layer 1 consists almost exclusively of dendrites serving the excitatory pyramidal cells in all other layers, but particularly 2, 3, and 5. Bottom-up signals normally exit the lower-level column from layer 2/3 and enter the higher column at level 4, which contains primarily excitatory stellate cells. These resemble pyramidals but tend to have shorter axons and only synapse locally, in layer 4 and with the layer-2/3 pyramidals, which then convey signals both upward to higher columns and down to layers 5 and 6 of the same column. Top-down signals move primarily from layers 5 and 6 of the upper column to both the thalamus and the layer-1 dendritic mats of the lower column, then through layer 2/3 and on to layers 5 and 6 for further transmission down the hierarchy.

Although outnumbered by excitatory cells by a ratio of approximately five to one in the cortex, inhibitory neurons also play a vital role in columnar behavior: they help sparsify the activity patterns of the column’s excitatory cells. The primary inhibitory neurons are basket and chandelier cells, which have short spatial ranges of influence of no more than 100 micrometers, whereas projections from excitatory neurons may extend several millimeters (Kandel, Schwartz, and Jessell 2000). Inhibitors often synapse directly on the somas or axon hillocks of excitatory neurons, thereby exerting a strong blocking effect. The presynaptic terminals of an inhibitor’s axons emit GABA, whose receptors on postsynaptic terminals have slower binding and release times than those found at receptors for the excitatory neurotransmitters such as glutamate and AMPA. Thus, whereas the excitation of a pyramidal cell lasts 15–20 milliseconds, its inhibition normally endures for 100–150 msec (Rodriguez, Whitson, and Granger 2004).

Lateral inhibition is commonplace in the brain, such that when a neuron N fires, it excites local inhibitors that quell the activity of neighboring neurons and (after 15–20 msec) N itself, all for the extended interval of 100–150 msec. This serves at least two purposes: (1) immediate hampering of neighbors promotes sparse activation patterns, which are much more convenient for information transmission and storage; and (2) strong, enduring inhibition of the originally active neuron(s), which constitute an activation pattern, gives other neurons an opportunity to participate in the next pattern, which helps reduce the overlap (and possible interference) between activation states, an obvious advantage for information storage and retrieval in a distributed memory.

The precise anatomical and physiological details of cortical columns (Mountcastle 1998; Schneider 2014; Rodriguez, Whitson, and Granger 2004) are far beyond the scope of this book, but some have special relevance for predictive coding, despite the lack of a fully comprehensive and empirically validated theory. In particular, the role of inhibition seems critical, since predictive coding entails a weakening of bottom-up signaling when top-down predictions match the rising sensory patterns. How could higher levels inhibit lower levels?

First of all, the short spatial range of inhibitory neurons indicates that level K+1



Download



Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.