The Seven Definitions of Learning

first:

  • **Learning is adjusting the parameters of an internal model through the prediction error. **
  • The human brain is a fantastic statistician and way better than any machine.

second:

  • 86 billion neurons with 10 000 synaptic connections each
  • the brain breaks down problems by creating a hirachie of multi-level models (like in listening to language)
    Denk-Hierachie

third:

  • each layer can only discover a small part of reality but together the layers in the brain form a very powerful learning device.
  • through the prediction error the brain always guesses what happens next and through a error the brain corrects its previous assumption to the right direction. Therefore every error delivers valuable information

forth:

  • randomly chosing parameters you can improve the robustness of learning

fifth:

  • Am I winning or losing? We constantly evaluate our chances of winning, so we also avoid actions which are more likely to make us lose
  • chess players train themselves by playing against themselves. As AlphaGo AI learns the best when playing against itself so the winning strategy strengthens while the losing strategy weakens.
  • we get better at discussing when someone else imitates our arguments
    Sind wir am Gewinnen

sixth:

  • We learn from everything we have seen or done and we use the learned stuff to predict new situations
    Prediction Error

seventh:

  • what I learned in one place can be generalised everywhere else
  • the more innate assumptions there are, the faster learning is
  • learning always starts from a set of priori hypotheses and the system selects those best suited to the current environment.
  • “To learn is to eliminate” - Jean-Pierre Changeux

see also

Tags: neuroscience science
Superlink: 050 🧠Neuroscience

Quellen

Chapter 1 Seven Defintions of Learning

Erstellt: 27-04-22 13:07