Pod A Conference Room
Working Group
This event is by invitation only.

Abstract.  Many problems that brains of essentially all species face involve attaching a neural tag to an arbitrary point in a high-dimensional space. A tag is a relatively small population of neurons that respond only to stimuli that represent a particular point in that high-dimensional space. For example, each of the 6 billion faces in the world occupies a distinct point in a very high-dimensional space (modern face recognition software uses 104 measurements on features of each face, so this software represents each face as a location in a 104 dimensional space). Humans have to very rapidly generate a (nearly) distinct tag for each face encountered so we can learn to recognize the faces that are significant to us. Another example of this same problem comes from olfaction: each odor represents a point in a high dimensional space, and we have to generate a tag that stands in for that odor in computations by the brain.

This problem is, perhaps, most easily studied in olfaction for 7 reasons. First, the same underlying mechanisms are conserved across many species from insects to humans. Second, we have a lot of detailed information about the anatomy and physiology of these olfactory systems. Third, we know the dimensionality of the space representing odors, and where it is located in the brain (for example, 50 dimensions for fruit fly located in the antennal lobe, 1000 dimensions for mouse, 350 dimensions for humans, both in the olfactory bulb). Fourth, in certain instances, we can know the location in the high-dimensional space that represents specific odors. Fifth, we know where and how the tags are generated (in fly, Kenyon cells in the mushroom body; in mammals, layer 2 semilunar cells in the anterior olfactory cortex). Sixth, in certain cases we can know exactly which neurons constitute the tag, and how the tag is used to identify the odor it represents. Seventh, we know basically how these tags are used in behavior (for example, spatial search for the source of an odor).

The goal of the workshop is to combine experimental data and theory to extend our understanding of this problem for olfaction, and to set the stage for understanding the same problem in other contexts.

SFI Host: 
Chuck Stevens