Showing papers on "Active learning (machine learning) published in 1994"

Programs for Machine Learning

Abstract: Algorithms for constructing decision trees are among the most well known and widely used of all machine learning methods. Among decision tree algorithms, J. Ross Quinlan's ID3 and its successor, C4.5, are probably the most popular in the machine learning community. These algorithms and variations on them have been the subject of numerous research papers since Quinlan introduced ID3. Until recently, most researchers looking for an introduction to decision trees turned to Quinlan's seminal 1986 Machine Learning journal article [Quinlan, 1986]. In his new book, C4.5: Programs for Machine Learning, Quinlan has put together a definitive, much needed description of his complete system, including the latest developments. As such, this book will be a welcome addition to the library of many researchers and students.

Neural Network Ensembles, Cross Validation, and Active Learning

Abstract: Learning of continuous valued functions using neural network ensembles (committees) can give improved accuracy, reliable estimation of the generalization error, and active learning. The ambiguity is defined as the variation of the output of ensemble members averaged over unlabeled data, so it quantifies the disagreement among the networks. It is discussed how to use the ambiguity in combination with cross-validation to give a reliable estimate of the ensemble generalization error, and how this type of ensemble cross-validation can sometimes improve performance. It is shown how to estimate the optimal weights of the ensemble members using unlabeled data. By a generalization of query by committee, it is finally shown how the ambiguity can be used to select new training data to be labeled in an active learning scheme.

An Introduction to Computational Learning Theory

Abstract: The probably approximately correct learning model Occam's razor the Vapnik-Chervonenkis dimension weak and strong learning learning in the presence of noise inherent unpredictability reducibility in PAC learning learning finite automata by experimentation appendix - some tools for probabilistic analysis.

WEKA: a machine learning workbench

Abstract: WEKA is a workbench for machine learning that is intended to aid in the application of machine learning techniques to a variety of real-world problems, in particular, those arising from agricultural and horticultural domains. Unlike other machine learning projects, the emphasis is on providing a working environment for the domain specialist rather than the machine learning expert. Lessons learned include the necessity of providing a wealth of interactive tools for data manipulation, result visualization, database linkage, and cross-validation and comparison of rule sets, to complement the basic machine learning tools. >

Robot juggling: implementation of memory-based learning

Abstract: Issues involved in implementing robot learning for a challenging dynamic task are explored in this article, using a case study from robot juggling. We use a memory-based local modeling approach (locally weighted regression) to represent a learned model of the task to be performed. Statistical tests are given to examine the uncertainty of a model, to optimize its prediction quality, and to deal with noisy and corrupted data. We develop an exploration algorithm that explicitly deals with prediction accuracy requirements during exploration. Using all these ingredients in combination with methods from optimal control, our robot achieves fast real-time learning of the task within 40 to 100 trials. >

MLC++: a machine learning library in C++

Abstract: We present MLC++, a library of C++ classes and tools for supervised machine learning. While MLC++ provides general learning algorithms that can be used by end users, the main objective is to provide researchers and experts with a wide variety of tools that can accelerate algorithm development, increase software reliability, provide comparison tools, and display information visually. More than just a collection of existing algorithms, MLC++ is can attempt to extract commonalities of algorithms and decompose them for a unified view that is simple, coherent, and extensible. In this paper we discuss the problems MLC++ aims to solve, the design of MLC++, and the current functionality. >

Reinforcement learning in continuous time: advantage updating

Abstract: A new algorithm for reinforcement learning, advantage updating, is described. Advantage updating is a direct learning technique; it does not require a model to be given or learned. It is incremental, requiring only a constant amount of calculation per time step, independent of the number of possible actions, possible outcomes from a given action, or number of states. Analysis and simulation indicate that advantage updating is applicable to reinforcement learning systems working in continuous time (or discrete time with small time steps) for which standard algorithms such as Q-learning are not applicable. Simulation results are presented indicating that for a simple linear quadratic regulator (LQR) problem, advantage updating learns more quickly than Q-learning by a factor of 100,000 when the time step is small. Even for large time steps, advantage updating is never slower than Q-learning, and advantage updating is more resistant to noise than is Q-learning. Convergence properties are discussed. It is proved that the learning rule for advantage updating converges to the optimal policy with probability one. >

Characterizing the applicability of classification algorithms using meta-level learning

Abstract: This paper is concerned with a comparative study of different machine learning, statistical and neural algorithms and an automatic analysis of test results. It is shown that machine learning methods themselves can be used in organizing this knowledge. Various datasets can be characterized using different statistical and information theoretic measures. These together with the test results can be used by a ML system to generate a set of rules which could also be altered or edited by the user. The system can be applied to a new dataset to provide the user with a set of recommendations concerning the suitability of different algorithms and these are graded by an appropriate information score. The experiments with the implemented system indicate that the method is viable and useful.

Learning one more thing

Abstract: Most research on machine learning has focused on scenarios in which a learner faces a single, isolated learning task. The lifelong learning framework assumes instead that the learner encounters a multitude of related learning tasks over its lifetime, providing the opportunity for the transfer of knowledge. This paper studies lifelong learning in the context of binary classification. It presents the invariance approach, in which knowledge is transferred via a learned model of the invariances of the domain. Results on learning to recognize objects from color images demonstrate superior generalization capabilities if invariances are learned and used to bias subsequent learning.

A review of machine learning in scheduling

Abstract: This paper has two primary purposes: to motivate the need for machine learning in scheduling systems and to survey work on machine learning in scheduling. In order to motivate the need for machine learning in scheduling, we briefly motivate the need for systems employing artificial intelligence methods for scheduling. This leads to a need for incorporating adaptive methods-learning. >

Learning to reason

Abstract: We introduce a new framework for the study of reasoning. The Learning (in order) to Reason approach developed here combines the interfaces to the world used by known learning models with the reasoning task and a performance criterion suitable for it. We show how previous results from learning theory and reasoning fit into this framework and illustrate the usefulness of the Learning to Reason approach by exhibiting new results that are not possible in the traditional setting. First, we give a Learning to Reason algorithm for a class of propositional languages for which there are no efficient reasoning algorithms, when represented as a traditional (formula-based) knowledge base. Second, we exhibit a Learning to Reason algorithm for a class of propositional languages that is not known to be learnable in the traditional sense.

Graph-based induction as a unified learning framework

Abstract: We describe a graph-based induction algorithm that extracts typical patterns from colored digraphs. The method is shown to be capable of solving a variety of learning problems by mapping the different learning problems into colored digraphs. The generality and scope of this method can be attributed to the expressiveness of the colored digraph representation, which allows a number of different learning problems to be solved by a single algorithm. We demonstrate the application of our method to two seemingly different learning tasks: inductive learning of classification rules, and learning macro rules for speeding up inference. We also show that the uniform treatment of these two learning tasks enables our method to solve complex learning problems such as the construction of hierarchical knowledge bases.

Instance-Based State Identification for Reinforcement Learning

Abstract: This paper presents instance-based state identification, an approach to reinforcement learning and hidden state that builds disambiguating amounts of short-term memory on-line, and also learns with an order of magnitude fewer training steps than several previous approaches. Inspired by a key similarity between learning with hidden state and learning in continuous geometrical spaces, this approach uses instance-based (or "memory-based") learning, a method that has worked well in continuous spaces.

Continuous action set learning automata for stochastic optimization

Abstract: The problem of optimization with noisy measurements is common in many areas of engineering. The only available information is the noise-corrupted value of the objective function at any chosen point in the parameter space. One well-known method for solving this problem is the stochastic approximation procedure. In this paper we consider an adaptive random search procedure, based on the reinforcement-learning paradigm. The learning model presented here generalizes the traditional model of a learning automaton [Narendra and Thathachar, Learning Automata: An Introduction, Prentice Hall, Englewood Cliffs, 1989]. This procedure requires a lesser number of function evaluations at each step compared to the stochastic approximation. The convergence properties of the algorithm are theoretically investigated. Simulation results are presented to show the efficacy of the learning method.

Simulation of Learning Development in Repetitive Construction

Abstract: This paper presents the methodology used to model the learning development phenomenon in the CYCLONE format using the Boeing learning curve. The learning development enhancement was coded in the MicroCYCLONE (a microcomputer version of CYCLONE) environment using QuickBASIC programming. The enhancement allows the user to specify the rate of learning expressed as a percentage, the realization count required for the task duration to increment due to learning, and the threshold at which no more learning improvement can be realized. The existing routines for generating deterministic durations and stochastic durations based on uniform, triangular, and beta distributions were upgraded to calculate durations that reflect the impact of learning development. The learning development enhancement can be extended for use with other simulation programs based on the CYCLONE modeling format. The developed learning‐development enhancement was used to model and simulate the learning‐development phenomenon for linear constr...

Ignoring data may be the only way to learn efficiently

Abstract: In designing learning algorithms it seems quite reasonable to construct them in a way such that all data the algorithm already has obtained are correctly and completely reflected in the hypothesis the algorithm outputs on these data. However, this approach may totally fail, i.e. it may lead to the unsolvability of the learning problem, or it may exclude any efficient solution of it. In particular, we present a natural learning problem and prove that it can be solved in polynomial time if and only if the algorithm is allowed to ignore data.

Active Learning with Statistical Models

Abstract: For many types of learners one can compute the statistically "optimal" way to select data. We review how these techniques have been used with feedforward neural networks [MacKay, 1992; Cohn, 1994]. We then show how the same principles may be used to select data for two alternative, statistically-based learning architectures: mixtures of Gaussians and locally weighted regression. While the techniques for neural networks are expensive and approximate, the techniques for mixtures of Gaussians and locally weighted regression are both efficient and accurate.

Active Learning for Function Approximation

Abstract: We develop a principled strategy to sample a function optimally for function approximation tasks within a Bayesian framework. Using ideas from optimal experiment design, we introduce an objective function (incorporating both bias and variance) to measure the degree of approximation, and the potential utility of the data points towards optimizing this objective. We show how the general strategy can be used to derive precise algorithms to select data for two cases: learning unit step functions and polynomial functions. In particular, we investigate whether such active algorithms can learn the target with fewer examples. We obtain theoretical and empirical results to suggest that this is the case.

A New Approach to Genetic Based Machine Learning and an Efficient Finding of Fuzzy Rules - Proposal of Nagoya Approach -

Abstract: This paper presents a new approach to genetic-based machine learning (GBML). The new approach utilizes mechanisms of genetic recombination in bacterial genetics, and the authors have called the new approach “Nagoya approach”. The Nagoya approach is efficient in improving local portions of chromosomes. An obstacle avoidance problem for a mobile robot is simulated using the Nagoya approach, and complex fuzzy rules are found.

An incremental learning algorithm that optimizes network size and sample size in one trial

Abstract: A constructive learning algorithm is described that builds a feedforward neural network with an optimal number of hidden units to balance convergence and generalization. The method starts with a small training set and a small network, and expands the training set incrementally after training. If the training does not converge, the network grows incrementally to increase its learning capacity. This process, called selective learning with flexible neural architectures (SELF), results in a construction of an optimal size network for learning all the given data using only a minimal subset of them. The author shows that the network size optimization combined with active example selection generalizes significantly better and converges faster than conventional methods. >

Iterative learning control for discrete-time nonlinear systems

Abstract: An iterative learning control method is proposed for a class of nonlinear discrete-time systems. The output of a system is shown to converge to a desired output for a finite time interval by using the proposed learning method under a certain condition. The proposed learning control method is applied to the tracking control of a two-link robot manipulator and the effectiveness of the learning is observed in the simulation

An on-line unsupervised learning machine for adaptive feature extraction

Abstract: Adaptive feature extraction is useful in many information processing systems. This paper proposes a learning machine implemented via a neural network to perform such a task using the tool principal component analysis. This machine (1) is adaptive to nonstationary input, (2) is based on an unsupervised learning concept and requires no knowledge of if, or when, the input changes statistically, and (3) performs online computation that requires little memory or data storage. Associated with this machine, the authors propose a learning algorithm (LEAP), whose convergence properties are theoretically analyzed and whose performance is evaluated via computer simulations. Two major contributions of this paper are: (1) Under appropriate conditions, the authors prove that the algorithm will extract multiple principal components, when the learning rate is constant; and (2) they identify a near optimal domain of attraction. >

A Guide to the Literature on Learning Graphical Models

Abstract: This literature review discusses different methods under the general rubric of learning Bayesian networks from data, and more generally, learning probabilistic graphical models. Because many problems in artificial intelligence, statistics and neural networks can be represented as a probabilistic graphical model, this area provides a unifying perspective on learning. This paper organizes the research in this area along methodological lines of increasing complexity.

Design and learning with cellular neural networks

Abstract: The template coefficients (weights) of a CNN, which will give a desired performance, can either be found by design or by learning; "By design" means, that the desired function to be performed could be translated into a set of local dynamic rules, while "by learning" is based exclusively on pairs of input and corresponding output signals, the relationship of which may be by far too complicated for the explicit formulation of local rules. An overview of design and learning methods applicable to CNNs, which sometimes are not clearly distinguishable,is given here. Both technological constraints imposed by specific hardware implementation and practical constraints caused by the specific application and system embedding are influencing design and learning. >

The Power of Self-Directed Learning

Abstract: This article studies self-directed learning, a variant of the on-line (or incremental) learning model in which the learner selects the presentation order for the instances. Alternatively, one can view this model as a variation of learning with membership queries in which the learner is only “charged” for membership queries for which it could not predict the outcome. We give tight bounds on the complexity of self-directed learning for the concept classes of monomials, monotone DNF formulas, and axis-parallel rectangles in l0, 1, …, n − 1rd. These results demonstrate that the number of mistakes under self-directed learning can be surprisingly small. We then show that learning complexity in the model of self-directed learning is less than that of all other commonly studied on-line and query learning models. Next we explore the relationship between the complexity of self-directed learning and the Vapnik-Chervonenkis (VC-)dimension. We show that, in general, the VC-dimension and the self-directed learning complexity are incomparable. However, for some special cases, we show that the VC-dimension gives a lower bound for the self-directed learning complexity. Finally, we explore a relationship between Mitchell's version space algorithm and the existence of self-directed learning algorithms that make few mistakes.

Genetic algorithms in noisy environment

Abstract: Genetic algorithms (GA) have been widely used in the areas of function optimization and machine learning. In many of these applications, the effect of noise is a critical factor in the performance of the genetic algorithms. In this paper, we propose an effective method for obtaining the optimal solution by using an optimal solution list and systematically changing certain parameters of the algorithm. Our results show that the optimal solution list is able to provide a small solution set that contains near optimal solutions. >