In many domains, an appropriate inductive bias is the MIN-FEATURES bias, which prefers consistent hypotheses definable over as few features as possible. This paper defines and studies this bias. First, it is shown that any learning algorithm implementing the MIN-FEATURES bias requires \Theta( 1 ffl ln 1 ffi + 1 ffl [2 p + p ln n]) training examples to guarantee PAC-learning a concept having p relevant features out of n available features. This bound is only logarithmic in the number of irrelevant features. The paper also presents a quasi-polynomial time algorithm, FOCUS, which implements MIN-FEATURES. Experimental studies are presented that compare FOCUS to the ID3 and FRINGE algorithms. These experiments show that--- contrary to expectations---these algorithms do not implement good approximations of MIN-FEATURES. The coverage, sample complexity, and generalization performance of FOCUS is substantially better than either ID3 or FRINGE on learning problems where the MIN-FEATURE...

Many real-world domains bless us with a wealth of attributes to use for learning. This blessing is often a curse: most inductive methods generalize worse given too many attributes than if given a good subset of those attributes. We examine this problem for two learning tasks taken from a calendar scheduling domain. We show that ID3/C4.5 generalizes poorly on these tasks if allowed to use all available attributes. We examine five greedy hillclimbing procedures that search for attribute sets that generalize well with ID3/C4.5. Experiments suggest hillclimbing in attribute space can yield substantial improvements in generalization performance. We present a caching scheme that makes attribute hillclimbing more practical computationally. We also compare the results of hillclimbing in attribute space with FOCUS and RELIEF on the two tasks. 1 INTRODUCTION As machine learning is applied to real-world tasks, difficulties arise that do not occur in simpler textbook experiments. One such diffic...

In most learning systems examples are represented as fixed-length "feature vectors", the components of which are either real numbers or nominal values. We propose an extension of the featurevector representation that allows the value of a feature to be a set of strings; for instance, to represent a small white and black dog with the nominal features size and species and the setvalued feature color, one might use a feature vector with size=small, species=canis-familiaris and color=fwhite,blackg. Since we make no assumptions about the number of possible set elements, this extension of the traditional feature-vector representation is closely connected to Blum's "infinite attribute" representation. We argue that many decision tree and rule learning algorithms can be easily extended to setvalued features. We also show by example that many real-world learning problems can be efficiently and naturally represented with set-valued features; in particular, text categorization problems and probl...

In this paper we study an extension of the distribution-free model of learning introduced by Valiant [23] (also known as the probably approximately correct or PAC model) that allows the presence of malicious errors in the examples given to a learning algorithm. Such errors are generated by an adversary with unbounded computational power and access to the entire history of the learning algorithm's computation. Thus, we study a worst-case model of errors. Our results include general methods for bounding the rate of error tolerable by any learning algorithm, efficient algorithms tolerating nontrivial rates of malicious errors, and equivalences between problems of learning with errors and standard combinatorial optimization problems. 1 Introduction In this paper, we study a practical extension to Valiant's distribution-free model of learning: the presence of errors (possibly maliciously generated by an adversary) in the sample data. The distribution-free model typically makes the idealize...

We analyze a few of the commonly used statistics based and machine learning algorithms for natural language disambiguation tasks and observe that they can be recast as learning linear separators in the feature space. Each of the methods makes a priori assumptions, which it employs, given the data, when searching for its hypothesis. Nevertheless, as we show, it searches a space that is as rich as the space of all linear separators. We use this to build an argument for a data driven approach which merely searches for a good linear separator in the feature space, without further assumptions on the domain or a specific problem. We present such an approach - a sparse network of linear separators, utilizing the Winnow learning algorithm - and show how to use it in a variety of ambiguity resolution problems. The learning approach presented is attribute-efficient and, therefore, appropriate for domains having very large number of attributes. In particular, we present an extensive experimental ...

We present a membership-query algorithm for efficiently learning DNF with respect to the uniform distribution. In fact, the algorithm properly learns with respect to uniform the class TOP of Boolean functions expressed as a majority vote over parity functions. We also describe extensions of this algorithm for learning DNF over certain nonuniform distributions and for learning a class of geometric concepts that generalizes DNF. Furthermore, we show that DNF is weakly learnable with respect to uniform from noisy examples. Our strong learning algorithm utilizes one of Freund's boosting techniques and relies on the fact that boosting does not require a completely distribution-independent weak learner. The boosted weak learner is a nonuniform extension of a parity-finding algorithm discovered by Goldreich and Levin. 3 1 Introduction Consider the following 20-questions-like game between two players, Bob and Alice. Bob has a Disjunctive Normal Form (DNF) expression f in mind. Alice is allo...

Abstract not found

This paper describes experimental results on using Winnow and Weighted-Majority based algorithms on a real-world calendar scheduling domain. These two algorithms have been highly studied in the theoretical machine learning literature. We show here that these algorithms can be quite competitive practically, outperforming the decision-tree approach currently in use in the Calendar Apprentice system in terms of both accuracy and speed. One of the contributions of this paper is a new variant on the Winnow algorithm (used in the experiments) that is especially suited to conditions with stringvalued classifications, and we give a theoretical analysis of its performance. In addition we show how Winnow can be applied to achieve a good accuracy/coverage tradeoff and explore issues that arise such as concept drift. We also provide an analysis of a policy for discarding predictors in Weighted-Majority that allows it to speed up as it learns. Keywords: Winnow, Weighted-Majority, Multiplicative alg...

We present an application of kernel methods to extracting relations from unstructured natural language sources.

Machine Learning research has been making great progress in many directions. This article summarizes four of these directions and discusses some current open problems. The four directions are (a) improving classification accuracy by learning ensembles of classifiers, (b) methods for scaling up supervised learning algorithms, (c) reinforcement learning, and (d) learning complex stochastic models.