Hinton diagrams: Viewing connection strengths in neural networks

A graphical analysis similar to that used by Hinton and Shallice (1991) was applied to the hair cells of a simulated auditory transducer. The graphical analysis made it apparent that there were hair cells that responded to a narrow range of frequencies, as would be predicted by the tonotopic organization of the real physiology. In short, this study demonstrates the efficacy of using graphic techniques to examine the nature of the autopoetic organization of the hidden layer of back-propagation networks.     

Measuring and predicting the effects of alcohol consumption on contrast sensitivity for stationary and moving gratings

Contrast sensitivity was measured for 12 healthy young males while sober, after ingestion of an alcohol placebo, and after ingestion of alcohol (95% grain alcohol; mean estimated blood alcohol level = .088%). Observations were made for both stationary gratings and gratings that traveled through a circular path and required pursuit eye movements. The significant alcohol-related reduction in contrast sensitivity was 2.6 times greater for moving (.29-log-unit reduction) than for stationary gratings (.11-log-unit reduction). The loss in contrast sensitivity for the moving gratings of high spatial frequency (12 cpd) was particularly severe (.37 log unit). Estimated blood alcohol level was correlated with the loss in contrast sensitivity for moving gratings (r = .61), but not with the loss for stationary gratings. Estimated blood alcohol level was strongly correlated with the difference between the loss in contrast sensitivity to moving and stationary gratings (r = .75). These results are consistent with reports that alcohol consumption degrades the ability to make pursuit eye movements. Subjects’ perceived intoxication level was not a reliable predictor of any index of visual performance.