Event

Taylor expansions for learning kernels and its potential applications to Speech

Oriol Vinyals

ICSI

Tuesday, August 19, 2008
12:30

Before joining ICSI, I worked on a Computer Vision project at CMU on how to learn kernel functions as Taylor expansions to improve kernel-based ML techniques. In that particular task, I focused on a database for pedestrian detection and used kernel SVMs. I will present the theoretical framework, some old results on the vision task, and recent work that I've done applying that technique to the Speaker ID task, showing some potential on other areas such as Speaker Diarization or Speech Recognition.

I will also give a brief description (high level) on what I did during my summer internship at Microsoft Research and how was the experience if there's time at the end of the talk.