Results 1  10
of
10
Budgeted learning of naivebayes classifiers
 IN PROCEEDINGS OF 19TH CONFERENCE ON UNCERTAINTY IN ARTIFICIAL INTELLIGENCE (UAI2003
, 2003
"... There is almost always a cost associated with acquiring training data. We consider the situation where the learner, with a fixed budget, may ‘purchase ’ data during training. In particular, we examine the case where observing the value of a feature of a training example has an associated cost, and t ..."
Abstract

Cited by 42 (3 self)
 Add to MetaCart
There is almost always a cost associated with acquiring training data. We consider the situation where the learner, with a fixed budget, may ‘purchase ’ data during training. In particular, we examine the case where observing the value of a feature of a training example has an associated cost, and the total cost of all feature values acquired during training must remain less than this fixed budget. This paper compares methods for sequentially choosing which feature value to purchase next, given the budget and user’s current knowledge of Naïve Bayes model parameters. Whereas active learning has traditionally focused on myopic (greedy) approaches and uniform/roundrobin policies for query selection, this paper shows that such methods are often suboptimal and presents a tractable method for incorporating knowledge of the budget in the information acquisition process.
Learning and Classifying under Hard Budgets
 In Proceedings of the European Conference on Machine Learning (ECML05
, 2005
"... Abstract. Since resources for data acquisition are seldom infinite, both learners and classifiers must act intelligently under hard budgets. In this paper, we consider problems in which feature values are unknown to both the learner and classifier, but can be acquired at a cost. Our goal is a learne ..."
Abstract

Cited by 32 (2 self)
 Add to MetaCart
Abstract. Since resources for data acquisition are seldom infinite, both learners and classifiers must act intelligently under hard budgets. In this paper, we consider problems in which feature values are unknown to both the learner and classifier, but can be acquired at a cost. Our goal is a learner that spends its fixed learning budget bL acquiring training data, to produce the most accurate “active classifier ” that spends at most bC per instance. To produce this fixedbudget classifier, the fixedbudget learner must sequentially decide which feature values to collect to learn the relevant information about the distribution. We explore several approaches the learner can take, including the standard “round robin” policy (purchasing every feature of every instance until the bL budget is exhausted). We demonstrate empirically that round robin is problematic (especially for small bL), and provide alternate learning strategies that achieve superior performance on a variety of datasets. 1
Budgeted learning of bounded active classifiers
 In Proceedings of the ACM SIGKDD Workshop on UtilityBased Data Mining
, 2005
"... Abstract. Since resources for data acquisition are seldom infinite, the need exists for learners and classifiers that act intelligently under hard budgets. In this paper, we consider problems in which feature values are unknown to the learner and classifier, but can be acquired at a cost. The goal i ..."
Abstract

Cited by 2 (1 self)
 Add to MetaCart
Abstract. Since resources for data acquisition are seldom infinite, the need exists for learners and classifiers that act intelligently under hard budgets. In this paper, we consider problems in which feature values are unknown to the learner and classifier, but can be acquired at a cost. The goal is a learner that spends its learning budget bL acquiring training data so as to produce the most accurate active classifier that spends at most bC per instance. From the learner’s perspective, purchasing every feature of every instance is sure to approach the underlying distribution asymptotically, but will this yield the best distribution when only bL dollars worth of data can be collected? In this work, we show empirically that the answer is no (especially for small bL) and present alternate learning strategies that achieve superior performance on a variety of realworld datasets. 1
Characterizing the Benefits of ModelBased Vs. DirectControl Learning in Exploration
"... : We consider the problem of learning a good control function from reinforcement. Although much studied, the issue of how (or even whether) to use prior knowledge to scale reinforcement learning up to realworld problems remains an important open issue. We investigate the inherent datacomplexity of ..."
Abstract
 Add to MetaCart
: We consider the problem of learning a good control function from reinforcement. Although much studied, the issue of how (or even whether) to use prior knowledge to scale reinforcement learning up to realworld problems remains an important open issue. We investigate the inherent datacomplexity of reinforcement learning when prior knowledge is expressed either as a set of possible reward models (modelbased learning) or as a set of possible controllers (directcontrol learning). Our key observation is that the distinction between these two approaches arises solely from the partial evaluation nature of the task (i.e. that each training instance evaluates only a subset of the controllers), and does not arise from temporal credit assignment or exploration/exploitation concerns per se. This allows us to compare these two learning approaches in a very simple batch setting. Our results show that it is not always beneficial to pursue a purely modelbased approach, even though this can yield...
Using Fewer Examples to Simulate Equivalence Queries
, 1996
"... It is well known that an algorithm that learns exactly using Equivalence queries can be transformed into a PAC algorithm that asks for random labelled examples. The first transformation due to Angluin (1988) uses a number of examples quadratic in the number of queries. Later, Littlestone (1989) and ..."
Abstract
 Add to MetaCart
It is well known that an algorithm that learns exactly using Equivalence queries can be transformed into a PAC algorithm that asks for random labelled examples. The first transformation due to Angluin (1988) uses a number of examples quadratic in the number of queries. Later, Littlestone (1989) and Schuurmans and Greiner (1995) gave transformations using linearly many examples. We present here another analysis of Littlestone's transformation which is both simpler and gives better leading constants. Our constants are still worse than Schuurmans and Greiner's, but while ours is a worstcase bound on the number of examples to achieve PAC learning, theirs is only an expected one. 1 Introduction Angluin's model of exact learning from queries [2] is one of the most frequently studied models of computational learning. The goal of a query learning algorithm is to infer an exact description of some unknown target concept by asking queries about it to some teacher. Department of Software (LS...
Budgeted Learning, Part II: The NaïveBayes Case
"... There is almost always a cost associated with acquiring training data. We consider the situation where the learner, with a xed budget, may `purchase' data during training. In particular, we examine the case where each feature value has an associated cost, and the total cost of all feature val ..."
Abstract
 Add to MetaCart
There is almost always a cost associated with acquiring training data. We consider the situation where the learner, with a xed budget, may `purchase' data during training. In particular, we examine the case where each feature value has an associated cost, and the total cost of all feature values acquired during training must remain less than this xed budget. This paper compares methods for choosing which feature value to purchase next, given the budget and user's current knowledge of the Nave Bayes parameters.
Massive Online Teaching to Bounded Learners ABSTRACT
"... We consider a model of teaching in which the learners are consistent and have bounded state, but are otherwise arbitrary. The teacher is noninteractive and “massively open”: the teacher broadcasts a sequence of examples of an arbitrary target concept, intended for every possible online learning al ..."
Abstract
 Add to MetaCart
We consider a model of teaching in which the learners are consistent and have bounded state, but are otherwise arbitrary. The teacher is noninteractive and “massively open”: the teacher broadcasts a sequence of examples of an arbitrary target concept, intended for every possible online learning algorithm to learn from. We focus on the problem of designing interesting teachers: efficient sequences of examples that lead all capable and consistent learners to learn concepts, regardless of the underlying algorithm used by the learner. We use two measures of teaching efficiency: the number of mistakes made by the worstcase learner, and the maximum length of the example sequence needed for the worstcase learner. Our results are summarized as follows: • Given a uniform random sequence of examples of an nbit concept function, learners (capable of consistently learning the concept) with s(n) bits of state are guaranteed to make only O(n · s(n)) mistakes and exactly learn the concept, with high probability. This theorem has interesting corollaries; for instance, every concept c has a sequence of examples can teach c to all capable consistent online learners implementable with s(n)size circuits, such that every learner makes only Õ(s(n)2) mistakes. That is, all resourcebounded algorithms capable of consistently learning a concept can be simultaneously taught that concept with few mistakes, on a single example sequence. We also show how to efficiently generate such a sequence of examples online: using Nisan’s pseudorandom generator, each example in the sequence can be generated with polynomialtime overhead per example, with an O(n · s(n))bit initial seed.
Supervisor
"... Permission is hereby granted to the University of Alberta Library to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. The author reserves all other publication and other rights in association with the copyright in the ..."
Abstract
 Add to MetaCart
Permission is hereby granted to the University of Alberta Library to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. The author reserves all other publication and other rights in association with the copyright in the thesis, and except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatever without the author’s prior written permission.