Smiljana Petrovic

RANDOM SUBSETS SUPPORT LEARNING A MIXTURE OF HEURISTICS

Problem solvers, both human and machine, have at their disposal many heuristics that may support effective search. The efficacy of these heuristics, however, varies with the problem class, and their mutual interactions may not be well understood. The long-term goal of our work is to learn how to select appropriately from among a large body of heuristics, and how to combine them into a mixture that works well on a specific class of problems. The principal result reported here is that randomly chosen subsets of heuristics can improve the identification of an appropriate mixture of heuristics. A self-supervised learner uses this method here to learn to solve constraint satisfaction problems quickly and effectively.

Learning a Mixture of Search Heuristics

This chapter describes programs that solve constraint satisfaction problems with multiple heuristics. It demonstrates the varied efficacy of individual constraint-solving metrics and the potential power available from a mixture of heuristics that references them. It describes a weighted-mixture decision process, and explains how one autonomous learner constructs its own labeled training examples from its search experience, and then learns a weighted mixture from them. Four new techniques are introduced to manage a large body of conflicting heuristics, and illustrated with empirical results.

Using Lexical Stress in Authorship Attribution of Historical Texts

This paper presents some early results from a comprehensive project, whose goal is to investigate the use of intonation and lexical stress in authorship attribution. We demonstrate how lexical stress patterns extracted from written text can be used to train a variety of machine learning algorithms to perform attribution of texts of unknown or disputed authorship. Specifically, we apply our methodology to a collection of 18

Full Restart Speeds Learning.

^\mathrm{th}

Attribution of 18th Century Political Writings Using Machine Learning

century American and British political writings, and demonstrate how combining lexical stress with other lexical features can significantly improve the attribution results.

Tailoring a Mixture of Search Heuristics

Thomas Paine was one of the most widely read and influential writers in the era of Democratic Revolutions in the eighteenth and early nineteenth centuries. He was a philosopher and a political leader who affected the course of the great revolutions in America and France, and whose writings still have relevance in modern politics. It is important to clarify the body of work that Paine produced. There has been speculation as to what actually were his writings, what has been left out of collections, and what has been mistakenly added to his works. For example, Philip Foner in his once authoritative Complete Writings of Thomas Paine, includes “An Occasional Letter on the Female Sex” and remarks, “Although there is evidence to prove that this article … was not written by Paine … it has been included in the present edition” because it may represent the sentiments of Paine.1 “African Slavery in America” was left out of Eric Foner’s Thomas Paine: Collected Writings, indicating some doubt to the attribution to Paine.2 Until an accurate collected works can be established, the scholarship on Paine will remain incomplete, and the public and academia will continue to misattribute quotes and opinions to Paine without historical support. The goal of our text analysis project is to address this deficiency.