PAC and Online learning via Mathematical Logic

Dr. Carlos Alfonso Ruiz Guido, (Escuela Bourbaki/Oxford University)
Resumen: The work of Vapnik around PAC- learnability has been rediscovered independently by some logicians (Shelah among others), since then deep connections have been found between those two approaches. Despite both of them being highly theoretical, in the side of Machine Learning, PAC is an important concept that has implications in the applied side of Machine Learning like SPV or Boosting. Less known is the work of Macintyre-Karpinski around deep learning and some applications could be found there. In this talk I will survey some of the more important results in this line of investigations and talk about some new plausible lines of investigation both in the applied and theoretical side.