Conclusions
The central question we ask is "Does inhibitory STDP improve the population code of MT, enabling the speed and direction of motion to be better read out by a linear decoder?" We ask this question in a particular context: of having designed a spiking motion processing model that is capable of estimating optic flow, even of small or low contrast stimuli. However, we expect the findings to be general to linear decoding of low SNR inputs into spiking neurons in networks with inhibition. In summary, we find that the answer is, yes, there is a benefit to inhibitory STDP, but only if the onset latencies of the feed-forward neurons are sufficiently staggered in time. Why? The initial volley of spikes engages the lateral connectivity which is capable of predicting the subsequent spikes, and balancing the excitatory drive with inhibition. If the spikes arrive too quickly, there is not enough time to engage the inhibitory machinery.