Timezone: »

 
Poster
Self-Paced Learning for Latent Variable Models
M. Pawan Kumar · Benjamin D Packer · Daphne Koller

Tue Dec 07 12:00 AM -- 12:00 AM (PST) @
in Poster Session »

Latent variable models are a powerful tool for addressing several tasks in machine learning. However, the algorithms for learning the parameters of latent variable models are prone to getting stuck in a bad local optimum. To alleviate this problem, we build on the intuition that, rather than considering all samples simultaneously, the algorithm should be presented with the training data in a meaningful order that facilitates learning. The order of the samples is determined by how easy they are. The main challenge is that often we are not provided with a readily computable measure of the easiness of samples. We address this issue by proposing a novel, iterative self-paced learning algorithm where each iteration simultaneously selects easy samples and learns a new parameter vector. The number of samples selected is governed by a weight that is annealed until the entire training data has been considered. We empirically demonstrate that the self-paced learning algorithm outperforms the state of the art method for learning a latent structural SVM on four applications: object localization, noun phrase coreference, motif finding and handwritten digit recognition.

[ PDF] 

Author Information

M. Pawan Kumar (Stanford University)
Benjamin D Packer (Stanford)
Daphne Koller (insitro)

Daphne Koller is the Rajeev Motwani Professor of Computer Science at Stanford University and the co-founder and co-CEO of Coursera, a social entrepreneurship company that works with the best universities to connect anyone around the world with the best education, for free. Coursera is the leading MOOC (Massive Open Online Course) platform, and has partnered with dozens of the world’s top universities to offer hundreds of courses in a broad range of disciplines to millions of students, spanning every country in the world. In her research life, she works in the area of machine learning and probabilistic modeling, with applications to systems biology and personalized medicine. She is the author of over 200 refereed publications in venues that span a range of disciplines, and has given over 15 keynote talks at major conferences. She is the recipient of many awards, which include the Presidential Early Career Award for Scientists and Engineers (PECASE), the MacArthur Foundation Fellowship, the ACM/Infosys award, and membership in the US National Academy of Engineering. She is also an award winning teacher, who pioneered in her Stanford class many of the ideas that underlie the Coursera user experience. She received her BSc and MSc from the Hebrew University of Jerusalem, and her PhD from Stanford in 1994.

More from the Same Authors