Decoding from populations of MT neurons during motion-discrimination

Abstract

Motion discrimination is a classic model system for probing computations and circuits underlying perceptual decisions. Despite a long history of studying the sensitivity of single neurons, little is known about how direction can be read out from the activity of neural populations. We recorded from ensembles of MT neurons while monkeys performed a motion-discrimination task. We compared the performance of a simple, neurally plausible, decoder to the psychophysical performance and to the sensitivity of single neurons. We found that the population was more accurate than the best single neurons and performed at least as well as the monkey at our task. We also found that the joint response patterns of neurons was not needed to compute the optimal weight pattern. MT populations were most sensitive to the stimulus immediately following motion onset, which corresponded to psychophysical weights of the monkeys. These results provide empirical groundwork for extending single neuron studies of perception to the population level.