search

alex-berman / sonotopy

Library for perceptually analyzing an acoustic signal in real time
GNU General Public License v3.0
9 stars 1 forks source link

dynamic adaptation #15

Open alex-berman opened 13 years ago

alex-berman commented 13 years ago

After the initial training phase, the map adapts much slower to changes in the input. Consequently, when the input changes drastically (e.g. when switching from one kind of tune to another) the map will be less responsive for a while, until the new soundscape has been learned.

To address this, one could replace the time-controlled adaptation with an error-driven adaptation. The amount of error (distance between input and winning model) will control the update and learning parameters, so that drastic changes in the input automatically "reset" the map. We can call this dynamic adaptation, or plasticity. See e.g. http://www.loria.fr/~rougier/research/DSOM.html

alex-berman commented 13 years ago

Tentative solution in 365c1bd7110eedf384634d2a54316b77fb104376