Form and formation of probabilistic sensory-motor circuits
- Author(s): Bouchard, Kristofer Eric
- Advisor(s): Brainard, Michael S
- Sabes, Philip N
- et al.
The majority of distinct motor (and sensory) events are encountered as temporally ordered sequences with rich probabilistic structure. The control and perception of behavior is supported by neurons with both sensory and motor fields. Hebbian plasticity is thought to shape neural circuits to represent the statistics of inputs and outputs. These observations about the brain and behavior suggest that Hebbian plasticity could engrain sequence probability in sensory-motor circuits. The Bengalese finch produces a learned song composed of probabilistically sequenced syllables. In songbirds, HVC is a sensory-motor circuit essential for the temporal/sequential control and perception of song. We used a combination of behavioral measurements, electrophysiological recordings, and computational modeling to investigate if and how HVC activity reflects the probability of produced sequences.
In Chapter 1 we investigated the Hebbian mechanisms that give rise to synapses representing sequence probabilities using neural network simulations. We show that Hebbian plasticity with balanced competitive and homogenizing forces can develop a synaptic representation of either the forward or backward transition probabilities of input sequences, depending on the site of synaptic competition. This demonstrates that both the forward and backward sequence probabilities can be encoded in sensory-motor circuits.
Chapter 2 tested whether and how HVC neurons exhibit predictive activity modulated by the forward probability of upcoming syllables by recording neural activity evoked during and following the playback of syllable sequences. We show that playback of a sequence induces post-stimulus activity that is similar to neural activity evoked by the next syllable in the sequence for low variability transitions, but not for high variability transitions. This demonstrates that the forward probabilities of produced sequences are reflected in induced HVC activity.
Chapter 3 tested whether and how auditory feedback statistics are encoded in HVC by playing back pseudo-randomly sequenced syllables from the bird's repertoire. We find that auditory responses to individual syllables integrate over several syllables and are positively modulated by the backward probability that previous sequences were produced during singing. This demonstrates that the backward probabilities of produced sequences are reflected in evoked HVC activity.