Learning handwriting by evolution: a conceptual framework for performance evaluation and tuning

In this paper we propose a method for evaluating the performance of an evolutionary learning system aimed at producing the optimal set of prototypes to be used by a handwriting recognition system. The trade-off between generalization and specialization embedded into any learning process is managed by iteratively estimating both consistency and completeness of the prototypes, and by using such an estimate for tuning the learning parameters in order to achieve the best performance with the smallest set of prototypes. Such estimation is based on a characterization of the behavior of the learning system, and is accomplished by means of three performance indices. Both the characterization and the indices do not depend on either the system implementation or the application, and therefore allow for a truly black-box approach to the performance evaluation of any evolutionary learning system.     

A risk-reward model for the on-line leasing of depreciable equipment

The optimal deterministic strategy for the on-line leasing of depreciable equipment is presented here for cases with and without an interest rate. A risk-reward model in which the on-line investor can develop optimal strategies based on his/her risk tolerance and forecast is discussed. Numerical analysis shows that the competitive performance is significantly improved in the proposed model.     

Machine Learning and Software Engineering

Machine learning deals with the issue of how to build programs that improve their performance at some task through experience. Machine learning algorithms have proven to be of great practical value in a variety of application domains. They are particularly useful for (a) poorly understood problem domains where little knowledge exists for the humans to develop effective algorithms; (b) domains where there are large databases containing valuable implicit regularities to be discovered; or (c) domains where programs must adapt to changing conditions. Not surprisingly, the field of software engineering turns out to be a fertile ground where many software development and maintenance tasks could be formulated as learning problems and approached in terms of learning algorithms. This paper deals with the subject of applying machine learning in software engineering. In the paper, we first provide the characteristics and applicability of some frequently utilized machine learning algorithms. We then summarize and analyze the existing work and discuss some general issues in this niche area. Finally we offer some guidelines on applying machine learning methods to software engineering tasks and use some software development and maintenance tasks as examples to show how they can be formulated as learning problems and approached in terms of learning algorithms.     

Finding frequent items over sliding windows with constant update time

In this paper, we consider the problem of finding ?-approximate frequent items over a sliding window of size N. A recent work by Lee and Ting (2006) [7] solves the problem by giving an algorithm that supports query and update time, and uses space. Their query time and memory usage are essentially optimal, but the update time is not. We give a new algorithm that supports O(1) update time with high probability while maintaining the query time and memory usage as .     

A note on the online First-Fit algorithm for coloring k-inductive graphs

In a FOCS 1990 paper, S. Irani proved that the First-Fit online algorithm for coloring a graph uses at most O(klogn) colors for k-inductive graphs. In this note we provide a very short proof of this fact.     

Learning with Maximum-Entropy Distributions

We are interested in distributions which are derived as a maximum entropy distribution from a given set of constraints. More specifically, we are interested in the case where the constraints are the expectation of individual and pairs of attributes. For such a given maximum entropy distribution (with some technical restrictions) we develop an efficient learning algorithm for read-once DNF. We extend our results to monotone read-k DNF following the techniques of (Hancock & Mansour, 1991).     

Identification of partial disjunction, parity, and threshold functions

Let F be a class of functions obtained by replacing some inputs of a Boolean function of a fixed type with some constants. The problem considered in this paper, which is called attribute efficient learning, is to identify “efficiently” a Boolean function g out of F by asking for the value of g at chosen inputs, where “efficiency” is measured in terms of the number of essential variables. We study the query complexity of attribute-efficient learning for three function classes that are, respectively, obtained from disjunction, parity, and threshold functions. In many cases, we obtain almost optimal upper and lower bound on the number of queries.     

On-Line Learning of Rectangles and Unions of Rectangles

Machine Learning - We design efficient algorithms for on-line learning of axis-parallel rectangles (and for the union of two such rectangles) in the common model for on-line learning with...     

A new measure for the study of on-line algorithms

An accepted measure for the performance of an on-line algorithm is the competitive ratio introduced by Sleator and Tarjan. This measure is well motivated and has led to the development of a mathematical theory for on-line algorithms.We investigate the behavior of this measure with respect to memory needs and benefits of lookahead and find some counterintuitive features. We present lower bounds on the size of memory devoted to recording the past. It is also observed that the competitive ratio reflects no improvement in the performance of an on-line algorithm due to any (finite) amount of lookahead.We offer an alternative measure that exhibits a different and, in some respects, more intuitive behavior. In particular, we demonstrate the use of our new measure by analyzing the tradeoff between the amortized cost of on-line algorithms for the paging problem and the amount of lookahead available to them. We also derive on-line algorithms for theK-server problem on any bounded metric space, which, relative to the new measure, are optimal among all on-line algorithms (up to a factor of 2) and are within a factor of 2K from the optimal off-line performance.     

On the power of randomization in on-line algorithms

Against in adaptive adversary, we show that the power of randomization in on-line algorithms is severely limited! We prove the existence of an efficient simulation of randomized on-line algorithms by deterministic ones, which is best possible in general. The proof of the upper bound is existential. We deal with the issue of computing the efficient deterministic algorithm, and show that this is possible in very general cases.A previous version of this paper appeared in the22nd ACM STOC Conference Proceedings. Part of this research was performed while A. Borodin and A. Wigderson were visitors at the International Computer Science Institute, and while G. Tardos was a visitor at the Hebrew University.     

Improved Rooftop Detection in Aerial Images with Machine Learning

In this paper, we examine the use of machine learning to improve a rooftop detection process, one step in a vision system that recognizes buildings in overhead imagery. We review the problem of analyzing aerial images and describe an existing system that detects buildings in such images. We briefly review four algorithms that we selected to improve rooftop detection. The data sets were highly skewed and the cost of mistakes differed between the classes, so we used ROC analysis to evaluate the methods under varying error costs. We report three experiments designed to illuminate facets of applying machine learning to the image analysis task. One investigated learning with all available images to determine the best performing method. Another focused on within-image learning, in which we derived training and testing data from the same image. A final experiment addressed between-image learning, in which training and testing sets came from different images. Results suggest that useful generalization occurred when training and testing on data derived from images differing in location and in aspect. They demonstrate that under most conditions, naive Bayes exceeded the accuracy of other methods and a handcrafted classifier, the solution currently used in the building detection system.     

A strongly competitive randomized paging algorithm

Thepaging problem is that of deciding which pages to keep in a memory ofk pages in order to minimize the number of page faults. We develop thepartitioning algorithm, a randomized on-line algorithm for the paging problem. We prove that its expected cost on any sequence of requests is within a factor ofH _k of optimum. (H _k is thekth harmonic number, which is about ln(k).) No on-line algorithm can perform better by this measure. Our result improves by a factor of two the best previous algorithm.Partial support for D. D. Sleator was provided by DARPA, ARPA Order 4976, Amendment 20, monitored by the Air Force Avionics Laboratory under Contract F33615-87-C-1499, and by the National Science Foundation under Grant CCR-8658139.     

A note on the k-Canadian Traveller Problem

We consider the online problem k-CTP, which is the problem to guide a vehicle from some site s to some site t on a road map given by a graph G=(V,E) in which up to k (unknown) edges are blocked by avalanches. An online algorithm learns from a blocked edge when reaching one of its endpoints. Thus, it might have to change its route to the target t up to k times. We show that no deterministic online algorithm can achieve a competitive ratio smaller than 2k+1 and give an easy algorithm which matches this lower bound. Furthermore, we show that randomization can not improve the competitive ratio substantially, by establishing a lower bound of k+1 for the competitivity of randomized online algorithms against an oblivious adversary.     

Using Genetic Algorithms for Concept Learning

In this article, we explore the use of genetic algorithms (GAs) as a key element in the design and implementation of robust concept learning systems. We describe and evaluate a GA-based system called GABIL that continually learns and refines concept classification rules from its interaction with the environment. The use of GAs is motivated by recent studies showing the effects of various forms of bias built into different concept learning systems, resulting in systems that perform well on certain concept classes (generally, those well matched to the biases) and poorly on others. By incorporating a GA as the underlying adaptive search mechanism, we are able to construct a concept learning system that has a simple, unified architecture with several important features. First, the system is surprisingly robust even with minimal bias. Second, the system can be easily extended to incorporate traditional forms of bias found in other concept learning systems. Finally, the architecture of the system encourages explicit representation of such biases and, as a result, provides for an important additional feature: the ability todynamically adjust system bias. The viability of this approach is illustrated by comparing the performance of GABIL with that of four other more traditional concept learners (AQ14, C4.5, ID5R, and IACL) on a variety of target concepts. We conclude with some observations about the merits of this approach and about possible extensions.     

Piecemeal Learning of an Unknown Environment

We introduce a new learning problem: learning a graph bypiecemeal search, in which the learner must return every so often to its starting point (for refueling, say). We present two linear-time piecemeal-search algorithms for learningcity-block graphs: grid graphs with rectangular obstacles.     

Learning with Genetic Algorithms: An Overview

Genetic algorithms represent a class of adaptive search techniques that have been intensively studied in recent years. Much of the interest in genetic algorithms is due to the fact that they provide a set of efficient domain-independent search heuristics which are a significant improvement over traditional weak methods without the need for incorporating highly domain-specific knowledge. There is now considerable evidence that genetic algorithms are useful for global function optimization and NP-hard problems. Recently, there has been a good deal of interest in using genetic algorithms for machine learning problems. This paper provides a brief overview of how one might use genetic algorithms as a key element in learning systems.     

Improved Bounds on Quantum Learning Algorithms

In this article we give several new results on the complexity of algorithms that learn Boolean functions from quantum queries and quantum examples.