FireStats is not installed in the database
Laryngitis homøopatiske midler køb propecia edderkop bid retsmidler

NSF Workshop on the Science and Engineering of Learning

Author:movellan @ July 26th, 2007 3 Comments

I just came from a really fun workshop organized by NSF on the Science and Engineering of Learning . Here is a bulleted lists of some ideas that I found particularly important.

  • An interesting scientific question for the science of learning: Why do we sleep and why is it important for learning.
  • Operating in non-stationary environments
  • Are there universal organizing principles of learning accross time scales and spatial scales: predict the future, infomax.
  • In Machine Learning we tend to focus on solving one problem at a time. But the brain has to solve many problems, and solving many problems at once may be easier than solving probles separately, due to synergies between many problems.
  • In the visual pathway from retina to IT we have a chain of about 10-15 synaptic layers. This may point to the importance of understanding why the chose has chosen “deep” multi-layered learning.
  • We already have robot appliances: the diswasher, the washing machine, the dryer. Problem is they solve the problem very well by having a very controlled environment. We need systems that work in less controlled conditions.
  • Should we focus on general purpose robots or on robots that may be special purpose but can operate in unconstrained environments.
  • Should we focus general purpose intelligence or a bag of tricks. Or should we attempt to understand the general principles behind a bag of tricks.
  • Patt Kuh;s’ work shows that between 6 -12 months babies change from “citizens of the world” to being
    “languagge specific listeners”, . For example before 6 months japanese infants can differentiate r/l but by 12 months they no longer can.

    One critical question is why, and how can we reproduce such effects with technology.

  • Social psychologists talk about The mere pressence effect . People do perform better when there is another human present.
  • Howard Nusbaum has some cool work on Synthetic speech learning
  • Machine Learning can help clarify Chomskian “poverty of the stimulus” arguments
  • It is imporant to define benchmarks task for autonomous learning system.
  • There are well known vocabulary standards for different ages. We could use
    them to choose key words and track progress when using social robots as teachers.
  • We will soon be able to fit 100 million transistors in the space of a single neuron.

    How do we learn to imitate?