Abstract
How does the cortex combine information from multiple sources? We tested several computational models against data from steady-state electroencephalography (EEG) experiments in humans, using periodic visual stimuli combined across either retinal location or eye-of-presentation. A model in which signals are raised to an exponent before being summed in both the numerator and the denominator of a gain control nonlinearity gave the best account of the data. This model also predicted the pattern of responses in a range of additional conditions accurately and with no free parameters, as well as predicting responses at harmonic and intermodulation frequencies between 1 and 30 Hz. We speculate that this model implements the optimal algorithm for combining multiple noisy inputs, in which responses are proportional to the weighted sum of both inputs. This suggests a novel purpose for cortical gain control: implementing optimal signal combination via mutual inhibition, perhaps explaining its ubiquity as a neural computation.