opening it up with Common Lisp

Favorite weblogs

Lisp Related

Bill Clementson

Finding Lisp

Lemonodor

Lispmeister.com

Planet Lisp

Politics

Orcinus

Talking Points Memo

This Modern World

Working for Change

Other home

Polliblog

Recent Readings

Book review: Darwinia
Reviewed: Friday, August 11, 2006

Summer reading: Spin
Reviewed: Saturday, August 5, 2006

Runner
Reviewed: Tuesday, July 18, 2006

the Omnivoire's Delimma
Reviewed: Wednesday, July 12, 2006

the Golem's Eye
Reviewed: Wednesday, May 31, 2006





tinderbox

Flexibility and Specificity in Infant Motor Skill Acquistion
Karen E. Adolph and Marion A. Eppler, 2002 , (Paper URL)
Tuesday, October 26, 2004

I recently heard a talk by Karen Adolph, a developmental psychologist at New York University. She talked about how balance is fundamental to learning motor skills in general and to learning how to sit, crawl, creep and walk in particular. She has performed dozens (if not hundreds!) of studies of infants and adults performing these tasks in a variety of situations: moving up and down slopes, across bridges with and without support, wearing weights, changing the texture of the surface or of the footwear. What she has found supports the theory that two very different learning systems are involved in all of this. One takes a long time to train but has good transfer (what is learned in one situation can be adapted quickly to other similar situations). The other can learn very quickly (in one or two trials) but is also very situationally specific.

The first learning system was researched heavily by Harry Harlow back in the 1950s. He called it learning set theory or learning to learn. His research focus was with monkeys but similar results have been found with humans as well. The gist of the experimental setup is that the monkey gets two things to choose from, one of which has a reward under it. The things might be two different shapes or two shapes with different colors or whatever. In each trial, the reward is always under the same shape. The first time the monkey goes through this experiment, it takes hundreds or even thousands (!) of trials before it learns that the reward is always under the circle (for example). It also takes a very long time for the monkey to learn the correct shape on the next trial, and the next. Eventually, however, something wonderful happens: the monkey understands that it can determine where the reward will be after a single trial -- if the raisin was under the triangle this time, keep picking the triangle. Otherwise, pick the other shape. The monkey has learned to learn and this learning transfers well to other similar situations. Adolph believes that balance is learned in a similar fashion. Infants explore constantly while their bodies and their environment change. They learn slowly to master each postural system (i.e., sitting, crawling, creeping or walking) but their new skills transfer to all kinds of movement situations.

The other learning style is associative learning. Here, we quickly learn to associate one thing or situation with another. As we all know, this happens very quickly. What is less obvious is that it doesn't transfer readily -- a fact learned in one situation does not come to mind readily in others even when these others may appear (usually on retrospect) to be very similar. Adolph has a experimental setup that demonstrates this very effectively. A 20 to 30 foot pathway is constructed with one 4 foot section replaced by a material with very different friction (e.g., teflon) or by a very spongy material (foam rubber) covered with fabric. In either case, the different section is visibly obvious. Infants will walk down the path and slip on the teflon or fall into the form on the first trial but quickly learn the difference and navigate the odd section on later trials. This doesn't sound all that odd, but wait.

Adolph also tests adults. They sign a consent form that describes the experimental setup: that they will be walking down a path that has a section replaced by different friction or material. Then, they walk down the path with the obviously different section and..., they fall down! This is hard to believe but I've seen the film. Equally incredible is that infants that have learned about the 'funny' section in one setup don't transfer this knowledge to other, similar, setups. For example, if they learn about the foam section when it is covered with blue fabric, this knowledge doesn't transfer when the foam is covered with red fabric! (I'm not sure about transfer with adults).

This is great research on a topic which is central to embodied learning. I'm really curious about computational models that fit into the learning set framework. Neural networks are an obvious candidate but I'm not aware of them fitting into the learning to learn category. If anyone knows any details, please drop me a line.


Home | About | Quotes | Recent | Archives

Copyright -- Gary Warren King, 2004 - 2006