COMPUTATIONAL STUDY ON SENSORY MOTOR INTEGRATION AND SENSORY SEQUENCE LEARNING IN SONGBIRD HVC
Open Access
- Author:
- Wang, Linli
- Graduate Program:
- Physics
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- August 25, 2009
- Committee Members:
- Dezhe Jin, Dissertation Advisor/Co-Advisor
Dezhe Jin, Committee Chair/Co-Chair
John C Collins, Committee Member
Gong Chen, Committee Member
Jayanth R Banavar, Committee Member - Keywords:
- computational
songbird
HVC
sequence
recognition
learning - Abstract:
- Intracellular recording shows HVCRA neurons in songbirds have strong selective response to the bird’s own song. This selectivity is especially sensitive to the sequential order of the syllables spanning over several hundred milliseconds. The firing pattern of HVCRA is sparse in the sense that each HVCRA neuron only fires at most once during one motif and has such precise timing that it always gives a burst of spikes lasting around 10 ms. Developed from a song production model using synfire chains, we incorporate the sensory input into the spike propagation process by coincident detection of sensory input from upstream and lateral input inside the synfire chain. The specific wiring structure from sensory input to the synfire chain embeds the sensory sequence which can be recognized by our model. This specific structure can be learned from initial all-to-all weak connections with a postsynaptic voltage dependent learning mechanism. The stability of this model is shown by the robust performances of both recognition and learning under different synaptic noises and various defects in training sequence. In order to handle repetitive inputs, Potassium A-current is induced into our model to check the current surge due to repetitive inputs. The working region and its effect on the sequence recognition of A-current are investigated and its limited role on the single compartment model is concluded. In order to construct a more biological model and leave more room for A-current, the two-compartment model of a single HVCRA neuron replaces the single compartment model of the integrate-fire neuron. A more popular learning mechanism (spike-timing dependent plasticity) is applied to induce the specific wiring from sensory neurons to synfire chain neurons. The sequence recognition and learning of this revised network are studied. The enhanced robustness is partly due to the enlargement of the coincident region by A-current.