Research Group Leader

Prof. Dr. Matthias Bethge
Prof. Dr. Matthias Bethge
Phone: +49 7071 29-89017
Fax: +49 7071 29-25015


Secretary: Heike König
Phone: +49 7071 29-89018
Fax: +49 7071 29-25015


Data Analysis and Models of Neural Coding and Computation (Winter Term 06/07)
Course Content
Conclusions from data are inevitably model dependent. This course provides an introduction into model-driven data analysis, addressing recurrent issues of neural coding and computation. I will explain `paradigmatic' examples, rather than proving theorems. In addition to the lectures there will be short homework problems based on MATLAB which allow one to get hands-on experience with the new tools presented. The content can be sketched as follows (i do not expected that we will actually manage to reach the end of the list):
1) Introduction to neural coding&computation: (a) Complex behavior from simple units (neural networks). (b) Identifying neural systems (supervised learning) (c) Sensory coding and the natural environment (unsupervised learning+source coding) (d) Reliable signal transmission with unreliable neurons (channel coding). (e) The dimension of time (neural dynamics).
2) Focus on system identification: pseudo-inverse, Gaussians & 2nd- order correlations, linear prediction, spike-triggered average, generalized linear systems (cascade models), spike triggered covariance, kernel regression, and Gaussian processes.
3) Focus on unsupervised feature extraction: eigenfunctions/ eigenspaces, PCA, Fourier analysis, whitening, ICA, oriented PCA, factor analysis, sparse coding, and mixture models.
4) Factors of rate coding accuracy for populations of neurons: tuning functions, noise models, read out, correlations, and other constraints.
5) Focus on time, neural dynamics and smoothly changing signals: rate vs temporal coding, conductance-based neurons, integrate-and-fire neurons, stimulus reconstruction from renewal processes, and more fancy stuff.

Course Objective
Course for graduate students and motivated undergrads who seek to get an overview of model-driven data analysis.

Matrix calculus and basic probability theory.

Suggested Reading
Diamantaras & Kung (1996). Principal Component Neural Networks.
Dayan & Abbott (2001). Theoretical Neuroscience.
MacKay (2005). InformationTheory, Inference, and Learning Algorithms.

Day, Time & Location
weekly - 2hours, preferably Thursday 10am

Thursday, Oct. 19, 10am,
MPI for Biol. Cybernetics, Dept. Empirical Inference, Seminar room
Last updated: Wednesday, 27.02.2013