This course builds on recent insights that in many cases the computation done by the nervous system can be described using techniques from system identification theory, probabilistic modelling (machine learning) and information theory. The aim will be to examine work on computation in nervous system in terms of these methods.

The background needed to successfully take this course is a good grounding in mathematics, particularly with regard to probability and statistics, vectors and matrices. The mathematical level required is similar to that which would be obtained by students who did not have significant difficulties with the courses Mathematics for Informatics 1-4 taken in the first two years of the Informatics undergraduate syllabus. The Neural Computation (NC) course is recommended as preparation. Many concepts in the second half come from the Probabilistic Modelling & Reasoning (PMR) course. It is recommended to take this course in parallel, or otherwise be familiar with these concepts. If in doubt on this consult the instructor.

Instructors:
Mark van Rossum and
Chris Williams .

Course Tutor: TBA

- Overview of relevant neurobiology
- Neural encoding
- Neural decoding
- Mathematics of Hebbian learning
- Models of early visual coding (sparse coding, ICA and beyond)
- Lateral Interactions and Feedback
- Object recognition

There will be two assessed assignments worth in total 25%. There will be an exam worth 75%.

Now online First assignment . Deadline Mon March 10th, 4pm.

Second assignment is now available,
deadline Mon March 31st, 4pm.

Code nipa2code.zip,
data ImagesTest.mat (26MB)
ImagesTrain.mat (51 MB).

A model solution is available.

Natural Image Statistics by Aapo Hyvärinen, Jarmo Hurri, and Patrik O. Hoyer. Full version online. This will be supplemented by papers from the literature.

Neural encoding: lecture slides (2x2) , lecture slides (single page)

Higher Order Statistics slides (single page), slides (4up),

Emergence of Simple-Cell Receptive Field Properties by Learning a Sparse Code for Natural Images, Olshausen BA, Field DJ Nature, 381: 607-609 (1996)

Sparse Coding with an Overcomplete Basis Set: A Strategy Employed by V1? Olshausen BA, Field DJ, Vision Research, 37: 3311-3325 (1997).

The "independent components" of natural scenes are edge filters. Anthony J. Bell and Terrence J. Sejnowski, Vision Research 37(23): 3327-3338 (1997). preprint, official paper also available electronically via UoE library (Science Direct)

A hierarchical Bayesian model for learning non-linear statistical regularities in non-stationary natural signals, Y. Karklin and M. S. Lewicki, Neural Computation, 17 (2): 397-423, (2005)

Learning higher-order structures in natural images, Y. Karklin and M. S. Lewicki, Network: Computation in Neural Systems, 14: 483-499, (2003)

Topographic Independent Component Analysis, A. Hyvarinen, P.O. Hoyer and M. Inki, Neural Computation, 13(7):1527-1558, (2001)

Emergence of phase and shift invariant features by decomposition of natural images into independent feature subspaces A. Hyvarinen and P.O. Hoyer, Neural Computation, 12(7):1705-1720, (2000)

Undirected Graphical Models lecture slides (single page), lecture slides (4up).

Natural Images, Gaussian Mixtures and Dead Leaves, Daniel Zoran and Yair Weiss, NIPS 2012.

Sparse deep belief net model for visual area V2. Honglak Lee, Ekanadham Chaitanya, and Andrew Y. Ng. Advances in Neural Information Processing Systems 20 (2008).

Topographic Product Models Applied To Natural Scene Statistics Osindero, S., Welling, M. and Hinton, G. E. Neural Computation, 18(2) (2006).

Natural Image Coding in V1: How Much Use is Orientation Selectivity? Eichhorn J, Sinz FH and Bethge M, PLoS Computational Biology 5(4:e1000336) 1-16 (2009)

Predictive Coding in the Visual Cortex. Rao, R. P. N. and Ballard, D. H., Nature Neuroscience, 2(1), 79-87, (1999)

Hierarchical Bayesian inference in the visual cortex. Lee, T.S., Mumford, D., Journal of Optical Society of America, A. . 20(7): 1434-1448 (2003)

Robust Object Recognition with Cortex-Like Mechanisms Serre, T., L. Wolf, S. Bilschi, M. Riesenhuber and T. Poggio. IEEE PAMI 29(3) 411-426 (2007)

How Close Are We to Understanding V1? Olshausen BA, Field DJ, Neural Computation, 17, 1665-1699 (2005).

20 years of learning about vision: Questions answered, questions unanswered, and questions not yet asked. Olshausen BA. In: 20 Years of Computational Neuroscience, Springer (2010)

*This page is maintained by
Chris Williams and Mark van Rossum *.

Informatics Forum, 10 Crichton Street, Edinburgh, EH8 9AB, Scotland, UK
Tel: +44 131 651 5661, Fax: +44 131 651 1426, E-mail: school-office@inf.ed.ac.uk Please contact our webadmin with any comments or corrections. Logging and Cookies Unless explicitly stated otherwise, all material is copyright © The University of Edinburgh |