Dissertation/Thesis Abstract

Rapid Learning with Stochastic Focus of Attention
by Pelossof, Raphael A., Ph.D., Columbia University, 2011, 85; 3451712
Abstract (Summary)

We present a method to determine when to stop the evaluation of a decision-making process. The method determines to stop the evaluation process when the result of the full evaluation is obvious. This trait is highly desirable for margin-based Machine Learning algorithms where a classifier traditionally evaluates all the features for every example. However, some examples are easier to classify than others, a phenomenon which is characterized by the event when most of the features agree on the class of an example. By stopping the feature evaluation when encountering an easy to classify example, a margin-based Machine Learning algorithm can achieve substantial reduction in running times.

To determine when to stop the feature evaluation, we develop a set of novel sequential tests, the Sequential Thresholded Sum Tests (STST). These tests stop the partial evaluation of the sum when the result of the full summation is guaranteed with high probability. By making different assumptions on the data and the features different tests arise. In general we look at the feature evaluation process as a random walk and apply different Brownian motion early stopping inequalities to determine when to stop the walk. From these inequalities we derive a family of stopping thresholds for sequential feature evaluations under different assumptions.

We demonstrate the effectiveness of the different STST by speeding up several Online Learning algorithms on synthetic and real data.

Indexing (document details)
Advisor: Ying, Zhiliang
Commitee:
School: Columbia University
School Location: United States -- New York
Source: DAI-B 72/06, Dissertation Abstracts International
Source Type: DISSERTATION
Subjects: Statistics, Artificial intelligence, Computer science
Keywords: Attentive learning, Focus of attention, Rapid learning, Sequential threshold sum tests
Publication Number: 3451712
ISBN: 9781124571485
Copyright © 2019 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy
ProQuest