Nishant Mehta | Learnability, Stability and Uniform Convergence

  • When Mar 10, 2016 from 11:00 AM to 01:00 PM (Europe/Amsterdam / UTC100)
  • Where Northern Hemisphere
  • Add event to calendar iCal

Learnability, Stability and Uniform Convergence (Shalev-Shwartz, Shamir, Srebro, and Sridharan)

http://www.jmlr.org/papers/volume11/shalev-shwartz10a/shalev-shwartz10a.pdf

The paper shows how a certain notion of algorithm stability is equivalent to learnability in Vapnik's general setting of the learning problem, for bounded losses. Beyond that, there are quite a few interesting results in the paper and interesting examples to discuss.

I'm no longer covering 'Train faster, generalize better...' for reasons I can discuss when we meet but would rather not have on this public record!