A General Framework for Learning Weighted Automata
Dr Borja de Balle Pigem ( McGill University )
- 12:00 1st September 2014 ( Trinity Term 2014 )Lecture Theatre B
Weighted automata provide a concise algebraic parametrization for functions from strings to real numbers. Thisclass contains many well-known examples like deterministic finite automata (DFA) -- where values are binary -- and hiddenMarkov models (HMM) -- where values represent probabilities of strings. In this talk I will present a general frameworkbased on weighted automata which can be used to tackle a wide variety of learning problems involving sequential data,including classification, density estimation, and sequence tagging. I will then show how recent spectral algorithms forlearning HMM and sequence tagging models can be derived naturally within this framework.
Speaker bio
Borja Balle is currently a postdoctoral fellow at McGill University, and prior to that he obtained his PhD from Universitat Politècnica de Catalunya (UPC) in 2013. His research interests lie on the intersection between automata theory and machine learning, in particular on applications of spectral learning techniques to natural language processing, grammatical inference, and reinforcement learning. He served as area chair for NIPS 2014, and his research has been recognized with a best paper award at EACL 2012, best student paper at ICGI 2012, and runner-up for best student paper at NIPS 2012.Share this:
