N. Lesica - The resolution and complexity of the neural code for speech
Before attempting to build models that can predict the responses of neuronal populations, it is useful to determine the complexity of the responses so that the model structure can be chosen accordingly. For example, if it is sufficient to predict only the modulations in spike rate of individual neurons over long timescales, then a relatively simple model can be used, whereas if precise prediction of the joint spike times of all neurons in the population is required, a more complex model may be needed. We investigated the resolution and complexity of the neural code for speech using the gerbil midbrain as a model system. We found that significant information was carried by precise spike timing with sub-millisecond precision. We also found that the interaction between signal and noise correlations played an important role, and that these interactions were well captured by a simple generalized linear model. These results suggest that while the neural code for speech is complex, it may be possible to model it using a relatively simple framework.