Description:

In this course, we focus on modeling the primate perceptual neocortex using probabilistic graphical models, including Bayesian and Markov networks, and extensions to model time and change such as hidden Markov models and dynamic Bayesian networks. We are particularly interested in exploring what is known about how primates respond to complicated visual stimuli in areas of cortex beyond the earliest stages of the ventral and dorsal visual pathways. Primary references are drawn from the literature in computational and cognitive neuroscience, machine learning, and other fields that bear on how biological and engineered systems make sense of the world.

The preliminary lectures will sketch out the relevant areas of computational and cognitive neuroscience and related work in on probabilistic models machine learning and statistics. Since we expect a range of backgrounds among the students the next few lectures will be based on the textbook¹ by Bear, Connors and Paradiso [1] and include a basic introduction to the anatomy and physiology of neurons in the central nervous system covering Chapters 1–9, the basis for learning and memory covering Chapters 23–-25 which will also draw on material from In Search of Memory by Eric Kandel [3], and the primate visual system covering Chapter 12 and also drawing on material from Eye, Brain and Vision by David Hubel [2] which is available on line.

The lectures will review what is known about visual processing in the retina, lateral geniculate nuclei and early ventral and dorsal visual pathways in primate cortex. This review will sample current computational models used to explain early visual processing including local contrast normalization in the retina, surround suppression in retina and lateral geniculate, sparse coding in striate cortex, etc. Next we will attempt to gain some insight into what we do not yet know about early visual processing and survey several papers that offer theories and supporting evidence for later-stage, higher-level processing in the inferotemporal cortex. We will also cover the various classes of so-called neurally plausible models including but not limited to those employing probabilistic graphical models.

The remainder of the class will consist of student presentations of papers selected from a collection of classic and recent papers — including selections from Probabilistic Models of the Brain edited by Rao, Olshausen and Lewicki [4] and papers published in Neural Information Processing (NIPS) —- that attempt to model human image classification and object recognition. Each of these papers will involve a computational model and the goal is to critique these papers using the dual criteria of relevance to our understanding of biological vision and progress on the problems of computer vision.

Students will be graded on their presentation, class participation, and a project to be determined in collaboration with the teaching staff. Projects will include replicating and evaluating existing computational models and implementing novel models that extend or combine the features of existing ones. Small interdisciplinary group projects are encouraged. The projects will be graded on the basis of an initial proposal which will be due around midterm and a final report and demonstration due during the exam period. There will be no traditional midterm or final exams. Several standard still-image and video data sets will be made available for experiments and students will be provided access to a number of relevant libraries written in Matlab and C.

Location and Time:

MW, 4:15-5:30pm, Gates 100

 

Staff:

Instructor: Thomas Dean

Email: tld [at] google [dot] com

Office hours: by appointment

 

Course Assistant: Andy L. Lin

Email: ydna [at] stanford [dot] edu

 

Textbooks:

Neuroscience: Exploring the Brain by Bear, Connors and Paradiso (optional)
The textbook is on reserve at the Lane Medical Library. An electronic copy is available here (Stanford IP may be required).

Grading:

– Class participation including presentation (20%)
– Project proposal due around midterm (20%)
– Project documentation and demonstration (60%)

References

[1]	Mark F. Bear, Barry Connors, and Michael Paradiso. Neuroscience: Exploring the Brain (Third Edition). Lippincott Williams & Wilkins, Baltimore, Maryland, 2006.
[2]	David H. Hubel. Eye, Brain and Vision (Scientific American Library, Number 22). W. H. Freeman and Company, 1995.
[3]	Eric R. Kandel. In Search of Memory: The Emergence of a New Scince of Mind. W. W. Norton, New York, NY, 2006.
[4]	R. P. N. Rao, B. A. Olshausen, and M. S. Lewicki. Probabilistic Models of the Brain: Perception and Neural Function. MIT Press, Cambridge, MA, 2002.