9.S912:   Vision and learning - computers and brains

Fall 2013

Class Times:	Friday 11:00-2:00 pm
Units:	3-0-9
Location:	46-5193
Instructors:	Shimon Ullman, Tomaso Poggio, Ethan Meyers
Office Hours:	By appointment
Email Contact :	emeyers@mit.edu
Previous Class:	FALL 2012

Course description

We will review and discuss research on the problem of learning to understand the world and interact with it using sensory information. We will use vision as the main domain, and review relevant learning approaches from both computational and biological perspectives. Topics will include learning in computational vision, recent advances and limitations, face processing by computers and brain, learning in synapses, reinforcement leaning and Markov decision process in computers and brains.

Class projects: The final assignment is a short paper, about 5 pages or so (due by December 11th). It needs to be related to the study of vision and learning, but other than that it's open. It can take one of several forms.

• First, it can be a topic that interests you, and can be related to your research. If you have a suggested topic, send Shimon Ullman an e-mail with a short description in order to get some feedback.
• A second option is to read a number of papers on a selected topic that you found interesting. If you have a preference, again you can e-mail to Shimon who can help with papers selection. The project should then be a summary of the papers (about 3 pages), and then a critique and suggestions for future studies (about 2 pages).
• Third, instead of a paper you can write a wikipedia entry related to vision and learning (again send Shimon and email about the topic you will cover and why the wikipedia entry should be updated)

Readings and video lectures

1. M-theory: Learning representations for learning like humans learn
   [1] Slides
   [2] Video lecture: Part 1 , Part 2
   


2. Sensory gating, integration and prediction
   [1] Video lecture: Part 1, Part 2
   


3. Learning in Spiking Networks
   [1] Perceptron Notes
   [2] Gutig R, Sompolinsky H (2005). The tempotron: a neuron that learns spike timing-based decisions. Nature Neuroscience 9, 42-428.
   [3] Gutig R, Sompolinsky H (2009). Time-Warp-Invariant Neuronal Processing. PLoS Biol 7(7).
   [4] Video lecture: Part 1, Part 2
   


4. Computational Reinforcement Learning
   [1] Reinfocement Learning: An Introduction Chapters 1, 3, and 6 are probably the most relevant.
   [2] Scholarpedia article on temporal difference learning
   [3] Recent developments in RL: Kober J, Bagnell JA, and Peters J (2013) Reinforcement Learning in Robotics: A Survey. Int. Journal of Robotics Research.
   [4] Lecture slides
   [5] Video lecture: Part 1, Part 2


5. Cortical processes for navigating complex acoustic environments
   [1] Video lecture: Part 1, Part 2


6. Computer Perception with Deep Learning
   [1] Video lecture: Part 1, Part 2
   [2] Lecture notes


7. Reinforcement learning in the brain
   [1] Niv Reinforcement learning in the brain
   [2] Video lecture: Part 1, Part 2


8. Machine Learning Approaches to Face Detection and Related Problems
   [1] P. Viola and M. Jones (2001). Rapid object detection using a boosted cascade of simple features CVPR.
   [2] R. Schapire and Y. Singer (1999). Improved boosting algorithms using confidence-rated predictions . Machine Learning.
   [3] Video lecture: Part 1, Part 2


9. The neural computations that underlie face processing in primates
   [1] Kanwisher N, McDermott J, & Chun M (1997). The Fusiform Face Area: A Module in Human Extrastriate Cortex Specialized for the Perception of Faces. Journal of Neuroscience. 17 4302-4311.
   [2] Freiwald WA, Tsao DY,( 2010). Functional compartmentalization and viewpoint generalization within the macaque face-processing system. Science Vol. 330 no. 6005 pp. 845-851.
   [3] Issa EB, DiCarlo JJ (2012). Precedence of the eye region in neural processing of faces. J. Neurosci. 32(47):16666-82.
   [4] Tan, C. and T. Poggio (2013). Faces as a Model Category" for Visual Object Recognition. MIT-CSAIL-TR-2013-004, CBCL-311, Massachusetts Institute of Technology, Cambridge, MA, March 18.
   [5] Video lecture: Winrich Freiwald and Elias Issa and Charles Cadieu
   


10. Using machine learning to understand biological vision and learning
    [1] S Chase, A Schwartz (2011) Inference from populations: going beyond modelsProgress in Brain Research, Vol. 192
    [2] F Tong, MS Prattle (2012) Decoding patterns of human brain activity. Annu Rev Psychol 64:483-509.
    [3] E Meyers, XL Qi, C Constantinidis (2012). Incorporation of new information into prefrontal cortical activity after learning working memory tasks. PNAS, 108:4651-4656.
    [4] Lecture slides

Prerequisites

The course is open to all graduate students; undergraduates can take the course with instructor permission.

Grading

Grading will be based on participation and a final project.

Schedule