Dealing with Missing Features in Learning Problems
Afshin Rostamizadeh (UC Berkly)
NICTA SML SEMINARDATE: 2011-04-14
TIME: 11:00:00 - 12:00:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
In this work, we introduce new online and batch algorithms that are robust to data with missing features, a situation that arises in many practical applications. For the batch learning scenario, we present a convex relation of a non-convex problem to jointly estimate an imputation function, used to fill in the values of missing features, along with the classification hypothesis. For the online learning scenario, we allow for the comparison hypothesis to change as a function of the subset of features that are observed on any given round, extending the standard setting where the comparison hypothesis is fixed throughout. We prove regret bounds in the online setting and Rademacher complexity bounds for the batch i.i.d. setting. The algorithms are tested on several UCI datasets, showing superior performance over baselines. This is joint work with Alekh Agarwal and Peter Bartlett.
BIO:
Afshin Rostamizadeh is currently a Post-doc in Peter Bartlett's group at UC Berkeley with research interests in machine learning theory and algorithms. He completed his PhD degree in Computer Science at the Courant Institute of Mathematical Sciences at New York University under the supervision of Mehryar Mohri, working on several problems related to learning with imperfect data, such as learning with biased training samples, learning with non-i.i.d. data and domain adaptation, as well as problems concerned with the design and analysis of algorithms used to select effective kernel functions for use within ubiquitous kernel-based methods.
