Notes for week 1

Barbara

Theme 1 (Jan 23) Introduction: The general issues and a little history

Since the 1950's scientists have been sticking electrodes in the brain and measuring the responses of single neurons to the outside world and in relation to behavior, mapping the connections between parts, removing and stimulating parts at macro and micro scales, in the hopes of figuring out how the brain works. Often deliberately, in hopes of bootstrapping their way to understanding, they did this without explicit guesses about what they might find, but as is typical they then carried unstated assumptions into the research.

Perception and representation

  1. The "single neuron doctrine." H. B. Barlow attempts to make the unstated assumptions explicit (Single units and sensation: A neuron doctrine for perceptual psychology?, Perception 1:371-394, 1972). He stated (correctly) that the general belief was that "features" of interest were represented by the maximal firing rate of neurons. Brain extracts features of interest (as determined by evolution or learning) and the rate of a neuron's measures the probability of the presence or absence of the feature.
  2. Extensions and problems:
    1. Does the dogma extend to ensembles of neurons? For example, in fMRI, the stated assumptions are: increase in metabolic rate = increase in neuronal firing = what an area of the brain codes. Example: fusiform face area.
    2. Are "features" being represented, or rather the statistically most advantageous representation of the (visual) information?
    3. Even primary sensory neurons vary their response over multiple dimensions (for example, location, intensity, onset and duration, type). What is being measured?
    4. Things not addressed in the original dogmas: integration of probability. Addition of dimension of neural response properties unfolding over time. Parallel and distributed representations with non-uniform computational properties.

Motor commands and the organization of action

  1. "Command neurons." A parallel entity to the feature extractor was imagined, the "command neuron." At various levels, the response rate of neuron represents and commands a desired action (change joint position; lunge for food; save money for college). Similarly, the concept and be extended to ensembles of neurons.
  2. Extensions and problems:
    1. Are perceptual and motor representations really symmetric?
    2. The problems of coming to decisions about the best course of action is highlighted in the motor system, where the relative value of a response must be integrated with calculations of the probability of its correctness.
    3. Action decisions make meaning explicit in the neuroanatomical connectivity flowing from the site of a decision — a neural "look-up table." How are hard-wired and plastic aspects of behavior resident in the same circuits managed?

[PowerPoint slides]

Shimon

General introductory remarks

Cognition as computation

Edelman's Challenge:
Pick a cognitive task, any task, and I'll explain how it is essentially and inherently computational.

Overheard in another context:

"Help me, Obi-Wan Kenobi. [...]"

The role of probability in cognition

From J. J. Gibson's review of a book by Egon Brunswick:
"The percept is always a wager. Thus uncertainty enters at two levels, not merely one: the configuration may or may not indicate an object, and the cue may or may not be utilized at its true indicative value." Gibson, J. J. (1957). Survival in a world of probable objects. Contemporary Psychology, 2:33-35.

Bayes Theorem

A few slides from Psych 214 introducing the Bayesian framework.

Bayes Networks

A few more slides (see esp. #11) introducing Bayesian networks (a kind of graphical model), with examples from various cognitive tasks.

Shimon Edelman <se37 at cornell.edu>
Last modified on Tue Jan 22 13:20:49 2008