Prof. Emo Welzl and Prof. Bernd Gärtner
|Mittagsseminar Talk Information|
Date and Time: Thursday, November 27, 2014, 12:15 pm
Duration: 30 minutes
Location: OAT S15/S16/S17
Speaker: Hafsteinn Einarsson
One shot learning is the task of retaining information presented only once. We study abstract models of one shot learning inspired by biology in the sense that we restrict ourselves to bioplausible parameter regimes. This study eventually led us to questions regarding the spread of information in random graphs (bootstrap percolation / iterative retrieval). For example, how many subsets with density p in a graph on n vertices can you insert until it fails to have the property that the activation of one subset stays within that subset and does not spread to the rest of the graph?
As it turns out, even though many of the models have asymptotically satisfying memory capacities, the restriction to bioplausible parameters can have a detrimental effect on the leading constants. This raises the question, what is the secret ingredient biology has which we are missing?
In this talk I will present past and ongoing research regarding one shot learning in an attempt to unfold this mystery.
Automatic MiSe System Software Version 1.4803M | admin login