A brief introduction to the sensory backdriving model

If I can do it justice with my own words...

The basic theory is that when we recall mental imagery, we activate many of the same neurons in primary visual cortex (V1) as when we actually experienced those same images. It follows then that since V1 fires as a direct topological map of what is being seen in the real world (great pictures here 12), then visualized images projected from higher cortex should also map out the same way in V1.

There is a fair amount of experimental evidence 1-5 supporting this theory, but it is also intuitively appealing. My two favorite reasons are:

Reason 1. At least as many axons run from higher processing regions of our brains toward V1 as run from V1 to higher processing. What is all this backtalk for? No mere camera gets this much feedback from its television monitor.

A simple explanation would be that the backtalk is doing a lot of pattern completion 4 for the seeing and interpretation of the real world. But where does pattern completion begin and imagination end? What about those times when no real data at all is coming in and we still fill in the gaps? We continue to visualize, and if the absence of sensory data is absolute and long enough, the visualizations become hallucinatory and indistinguishable from real sensory input. Pattern completion, visualization, and seeing seem to be part of the same thing.

Another explaination for the top-down connections is that they exist to direct attention. If that is true then at least one study 5 suggests that attention is not to a single point, but may be almost as spatially complex as visualization (which would make sense considering the bandwidth of the top-down connections). Therefore if attention projects such a detailed image upon visual cortex, it will serve the same purpose for our goals as visualization. I am not sure visualization and attention are separable anyway. Is it possible to imagine something without paying attention to it? How easy is it to anticipate seeing something without also imagining it? One is at least an indicator of the other.

Reason 2. Even if sensory backdriving is neither sufficient nor necessary for visualization, classical conditioning would make it an almost unavoidable side effect of living:

Conditioning in a Loop

Classical Conditioning of V1 in a Loop

(True, Grandma's image precedes the concept, but unless Grandma leaves the room before you recognize her, the concept will also precede the image.)

Classical conditioning would describe Path 1 (visual input) as the Unconditioned Stimulus to V1, and Path 3 (feedback) as the Conditioned Stimulus, so that after many repetitions, even when V1 receives no sensory input it will fire in the “Grandma” visual pattern when it receives “Grandma” concept-related firing patterns via Path 3.

...And even assuming the patterns coming down Path 3 contain only incidental leakage from conceptual activity - if they have any idiosyncrasies at all unique to the identification of “Grandma” then they will have a unique association with the sensory image of Grandma.

There is a lot more that can be said about the backdriving model. But the point is, if it is accurate, we might very well decode mental imagery if we can map out internally-generated V1 activity independently from visual input. On the next page, we'll cover a possible way to do that non-invasively.




Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 2.5 License.

September 30 2005