Title: Spectral Learning Methods for Probabilistic Automata

Abstract:
The goal of this talk is to present the spectral learning method
developed recently for Hidden Markov Models in the setting of
Probabilistic Finite Automata. We will begin by motivating the problem
from both the Machine Learning and Grammatical Inference perspectives.
Then the basic algorithm will be derived in detail and a proof sketch
of its sample complexity will be given. Finally, we will present an
extension of the basic algorithm capable of learning Probabilistic
Transducers.