Connectionism and wavelets in the modeling and analysis of neural system dynamics

grantor: University of TorontoThis thesis explores the analysis and modeling of sensori-motor systems in particular and neural systems in general. 'Chapter 1: Neural Networks and Motor Control' reviews the state of connectionist theory and demonstrates the importance of dynamics and biological relevance through a detailed case study. To this end, Chapters 2 and 3 document the development of novel approaches in the acquisition, analysis and modeling of dynamical non-linear neural processes. Chapter 2 augments current techniques in spectral analysis and introduces a calibrated version of the wavelet transform. The improved spectral and temporal resolutions of the algorithm, as compared with traditional Fourier-based joint time-frequency methods, is demonstrated through testing on synthetic data. The thesis then proceeds to describe four applications of the technique including: (i) Identification of EOG saccades; (ii) A first report of high-frequency transients in Kindling; (iii) Wavelet analysis of mu rhythms in EEG; (iv) The first application of wavelet analysis for the purpose of characterizing the dynamics of a recurrent neural network model. The final chapter draws technical and theoretical conclusions from the above applications. The thesis includes the source code for the data acquisition, wavelet analysis and modeling programs.M.Sc