Transcript Lecture Notes

Bayesian Brain
Presented by Nguyen Duc Thang
Contents




Introduction
Bottom-up approach
Top-down approach
Vision recognition, brain computer interface
(BCI), and artificial general intelligence (AGI)
Introduction

Old dream of all
philosophers and more
recently of AI:


understand how the brain
works
make intelligent machines
T. Poggio “Visual recognition in primates and machines”, NIPS’07 tutorial
Bayes rule
K. Kording “Decision Theory: What "Should" the Nervous System Do?”, Science 26 Oct. 2007
Bayes rule
Free energy and brain
Any adaptive change in the brain will minimize the freeenergy, this is correspondent to Bayesian inference
process: make prediction about the world and update
based on what it senses
Friston K., Stephan KE. “Free energy and the brain”, Synthese, 2007
Two approaches of Bayesian brain

Bottom-up approach


Top-down approach


How the brain works?
Machine intelligence
When two approaches meet together?
Bottom-up approach
Bayesian population code
- Single neural: the spike
counts satisfy the Poisson
distribution
- A group of neural: decode
the stimulus by Gaussian
distribution
Ma W.J.,Beck J., Latham P., Pouget A. “Bayesian inference with probabilistic population codes”,
Nature Neuroscience, 2006
Bayesian inference
Sum of two population
codes is equivalent to
taking the product of their
encoded distributions
Beck J., Ma W.J., Kiani R., Hanks T., Churchland A.K., Roitman L. , Shadlen M.N., Latham P.,
Pouget A. “Probabilistic population codes for Bayesian decision making ”, Neuron, 2008
Blue brain project
Top-down approach




Machine intelligence
Is based on the Bayes theorem, build a
probabilistic framework for one specific
problem, and apply Bayesian inference to
find solutions
Bayesian inference: belief propagation,
variational method, and non-parametric
method
Some journals: IJCV, PAMI, CVIU, JMLR
Interesting results
Automatically discover structure form,
ontology, causal relationships
Kemp C., Tenenbaum J. B. “The discovery of structural form”, PNAS 2008
Related researches



Vision recognition
Brain computer interface (BCI)
Artificial general intelligence (AGI)
David Hunter Hubel (born February 27, 1926) was co-recipient with Torsten Wiesel of the 1981 Nobel Prize in
Physiology or Medicine, for their discoveries concerning information processing in the visual system
Vision recognition
Classify animal and non-animal
Results
Serre T., Oliva A., Poggio T. “A feedforward architecture accounts for rapid categorization”, PNAS
2007
What is next: beyond the feedforward
models
Hierarchy Bayesian inference
Brain-Computer interface (BCI)




A brain–computer interface (BCI), sometimes called a direct neural interface or a
brain–machine interface, is a direct communication pathway between a brain and an
external devices
Invasive BCI: direct brain implants restore sight for blindness, hand-control for
persons with paralysis
Non-invasive BCI: EEG, MEG, MRI
Interesting results: research developed in the Advanced Telecommunications (ATR)
Computational Neuroscience LAB in Kyoto, Japan allowed the scientists to
reconstruct images directly from the brain and display them on a computer.
Miyawaki Y., “Decoding the mind’s eye-visual image reconstruction from human brain activity using
a combination of multiscale local image decoders”, Neuron Dec.2008
Artificial General Intelligence (Strong AI)




Weak AI only claims that machines can act
intelligently. Strong AI claims that a machine that
acts intelligently also has mind and understands
in the same sense people do
More information on the AGI conference 2009
Prediction: singularity in 2045
Two different opinions


I, robot (2004) Eagle eye (2008)
Cyborg girl (2008) Doraemon
My opinion