2017年第二届曲阜视觉科学会议知识库

Many medial superior temporal (MST) neurons are tuned to heading direction of self-motion based on optic flow or vestibular inputs. The preferred directions from different senses within a single neuron can be congruent (matched) or opposite from each other (Gu, Angelaki, DeAngelis 2008). Similarly, neurons in  middle temporal (MT) cortex are tuned to depth based on binocular disparity or motion parallax, and the preferred depths from different modalites can also be congruent or opposite (Nadler et al 2013). These are examples of neuron in the brain that respond to inputs from multiple sensory modalities, e.g., vision, audition, and touch,  or from multiple unisensory cues.  While the congruently tuned neurons appear natural for the functional role of cue integration, the oppositely tuned neurons appear somewhat puzzling.  I propose that, together, congruent and opposite cells in a neural population achieve efficient coding given incomplete redundancy in the input sources (Barlow, 1961). Efficiency requires creating representations (also called bases) in which the inputs are decorrelated. For two sources, this implies two bases in which inputs that sample the features from the two sources are either (weighted and) added or (weighted and) subtracted; these are the genesis respectively of congruently and oppositely tuned cells. 
The opposite cells are needed because the two sources are typically only incompletely redundant, and their activities can encode input information not encoded by the congruent cells. The exact forms (i.e., relative weighting of the sources) of, and neural sensitivities to, individual bases should depend on, and adapt to, the statistical properties of the inputs (e.g., the correlation between the sources and the signal to noise ratios). Coding of visual-vestibular heading direction and stereoscopic-motion-parallax depth both become analogous to efficient stereo coding (Li & Atick 1994). Generalization to more than two senses or modalities is straightforward. For example, coding of inputs from three sources can be analogous to efficient color coding of inputs from three cone types (Atick, Li, Redlich 1992).