Showing papers on "Active learning (machine learning) published in 1988"

Learning Quickly When Irrelevant Attributes Abound: A New Linear-Threshold Algorithm

Abstract: Valiant (1984) and others have studied the problem of learning various classes of Boolean functions from examples. Here we discuss incremental learning of these functions. We consider a setting in which the learner responds to each example according to a current hypothesis. Then the learner updates the hypothesis, if necessary, based on the correct classification of the example. One natural measure of the quality of learning in this setting is the number of mistakes the learner makes. For suitable classes of functions, learning algorithms are available that make a bounded number of mistakes, with the bound independent of the number of examples seen by the learner. We present one such algorithm that learns disjunctive Boolean functions, along with variants for learning other classes of Boolean functions. The basic method can be expressed as a linear-threshold algorithm. A primary advantage of this algorithm is that the number of mistakes grows only logarithmically with the number of irrelevant attributes in the examples. At the same time, the algorithm is computationally efficient in both time and space.

An empirical study of learning speed in back-propagation networks

Abstract: Most connectionist or "neural network" learning systems use some form of the back-propagation algorithm. However, back-propagation learning is too slow for many applications, and it scales up poorly as tasks become larger and more complex. The factors governing learning speed are poorly understood. I have begun a systematic, empirical study of learning speed in backprop-like algorithms, measured against a variety of benchmark problems. The goal is twofold: to develop faster learning algorithms and to contribute to the development of a methodology that will be of value in future studies of this kind. This paper is a progress report describing the results obtained during the first six months of this study. To date I have looked only at a limited set of benchmark problems, but the results on these are encouraging: I have developed a new learning algorithm that is faster than standard backprop by an order of magnitude or more and that appears to scale up very well as the problem size increases. This research was sponsored in part by the National Science Foundation under Contract Number EET-8716324 and by the Defense Advanced Research Projects Agency (DOD), ARPA Order No. 4976 under Contract F33615-87C-1499 and monitored by the Avionics Laboratory, Air Force Wright Aeronautical Laboratories, Aeronautical Systems Division (AFSC), Wright-Patterson AFB, OH 45433-6543. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of these agencies or of the U.S. Government.

Efficient Parallel Learning Algorithms for Neural Networks

Abstract: Parallelizable optimization techniques are applied to the problem of learning in feedforward neural networks. In addition to having superior convergence properties, optimization techniques such as the Polak-Ribiere method are also significantly more efficient than the Backpropagation algorithm. These results are based on experiments performed on small boolean learning problems and the noisy real-valued learning problem of hand-written character recognition.

An algebraic projection analysis for optimal hidden units size and learning rates in back-propagation learning

Abstract: Two critical issues in back-propagation (BP) learning are the discrimination capability given a number of hidden units and the speed of convergence in learning. The number of hidden units must be sufficient to provide the discriminating capability required by the given application. On the other hand, the training of an excessively large number of synaptic weights may be computationally costly and unreliable. This makes it desirable to have an a priori estimate of an optimal number of hidden neurons. Another closely related issue is the learning rate of the BP rule. In general, it is desirable to have fast learning, but not so fast that brings about instability of the iterative computation. An algebraic projection (AP) analysis method is proposed that provides an analytical solution to both of these problems. If the training patterns are completely irregular, then the predicted optimal number of hidden neurons is the same as that of the training patterns. In the case of regularity embedded patterns, the number of hidden neurons will depend on the type of regularity inherent. The optimal learning rate parameter is found to be inversely proportional to the number of hidden neurons. >

Machine Learning as an Experimental Science

Abstract: Machine learning is a scientific discipline and, like the fields of AI and computer science, has both theoretical and empirical aspects. Although recent progress has occurred on the theoretical front (see Machine Learning, volume 2, number 4), most learning algorithms are too complex for formal analysis. Thus, the field promises to have a significant empirical component for the foreseeable future. And unlike some empirical sciences, machine learning is fortunate enough to have experimental control over a wide range of factors, making it more akin to physics and chemistry than astronomy or sociology.

Task-level robot learning

Abstract: The functionality of robots can be improved by programming them to learn tasks from practice. Task-level learning can compensate for the structural modeling errors of the robot's lower-level control systems and can speed up the learning process by reducing the degrees of freedom of the models to be learned. The authors demonstrate two general learning procedures-fixed-model learning and refined-model learning-on a ball-throwing robot system. Both learning approaches refine the task command based on the performance error of the system, while they ignore the intermediate variables separation the lower-level systems. The authors also provide experimental and theoretical evidence that task-level learning can improve the functionality of robots. >

Generalizing number and learning from multiple examples in explanation based learning

Abstract: Explanation-based learning (EBL) systems have established their applicability to a wide variety of tasks. However, in despite intensive research, several problems relating to explanation-based learning have remained by and large open. This paper describes an approach to the problems of generalizing number and learning efficiently from multiple examples. The basic insight upon which the technique is based is that EBL can be thought of as learning control knowledge for a theorem-prover. By providing a richer representation for such control knowledge, more general rules can be learned: in particular, by providing looping constructs, rules which generalize number can be expressed; and by providing conditional branches, rules learned from different training examples can be combined. The technique described has been fully implemented, is domain-independent, and has been applied to a number of examples from the domain of VLSI circuit design.

Refinement of robot motor skills through reinforcement learning

Abstract: An extension of earlier work in the refinement of robotic motor control using reinforcement learning is described. It is no longer assumed that the magnitude of the state-dependent nonlinear torque is known. The learning controller learns about not only the presence of the torque, but also its magnitude. The ability of the learning system to learn this real-valued mapping from output feedback and reference input to control signal is facilitated by a stochastic algorithm that uses reinforcement feedback. A learning controller that can learn nonlinear mappings holds many possibilities for extending existing adaptive control research. >

A formal analysis of machine learning systems for knowledge acquisition

Abstract: Machine learning techniques can be of great value for automating certain aspects of knowledge acquisition. Given the potential of machine learning for knowledge acquisition, we have begun a systematic investigation of how one might map the functions of knowledge-based systems onto those machine learning systems that provide the required knowledge. The goal of our current research is to provide a general characterization of machine learning systems and their respective application domains.

Hypothesis Filtering: A Practical Approach to Reliable Learning

Abstract: This paper shows how to take virtually any learning algorithm and turn it into one which is accurate and reliable to an arbitrary degree The transformation is accomplished by appending a filter that performs a statistical test to the learning algorithm The filter only outputs hypotheses that pass the test The test takes time which is polynomial in the desired accuracy and reliability levels and is independent of the learning algorithm and the complexity of its domain Distribution-free statistical theory is used to prove that the filter works We describe the application of the filter to concept learning, and to SE, a system which learns control knowledge by clustering its data The significance of hypothesis filtering is two fold First, filters may be used to evaluate and compare the performance of a wide range of learning algorithms Second, applying hypothesis filtering to inductive learning algorithms demonstrates that the algorithms can be made arbitrarily reliable, and provably so

Creation by refinement: a creativity paradigm for gradient descent learning networks

Abstract: The author describes a paradigm for creating novel examples from the class of patterns recognized by a trained gradient-descent associative learning network. The paradigm consists of a learning phase, in which the network learns to identify patterns of the desired class, followed by a simple synthesis algorithm, in which a haphazard 'creation' is refined by a gradient-descent search complementary to the one used in learning. This paradigm is an alternative to one in which novel patterns are obtained by applying novel inputs to a learned mapping, and can be used for creative problems, such as music composition, which are not described by an input-output mapping. A simple simulation is shown in which a back-propagation network learns to judge simple patterns representing musical motifs, and then creates similar motifs. >

A Heuristic - Statistical Feature Selection Criterion For Inductive Machine Learning In The Real World

Abstract: Many machine learning methods have been developed for constructing decision trees from collections of examples. When they are applied to complicated real-world problems they often suffer from difficulties of coping with multi-valued or continuous features and noisy or conflicting data. To cope with these difficulties, a key issue is a powerful feature- selection criterion. After a brief review of the main existing criteria, this paper proposes a heuristic-statistical criterion symmetrical /spl tau/. This overcomes a number of the weaknesses of previous feature selection methods. Illustrative examples are presented.

Learning heuristic functions for numeric optimization problems

Abstract: Provides a framework for approaching the learning of heuristic functions for numeric optimization problem solutions. The outline of a system called 'Techniques using Experimentation for Acquisition and Creation of HeuRistics 2.0' (TEACHER 2.0), for learning heuristic functions is presented. The system is unique in that it combines many learning techniques into one coherent system. It can be a powerful learning system as it allows the generation of heuristics based on an amalgamation of learning techniques and strategies. The value of the system is illustrated by an example in which TEACHER 2.0 learns a new heuristic that is superior to the typical heuristic for that problem domain. >

Solutions to a class of nonstandard stochastic control problems with active learning

Abstract: The author formulates and solves a dynamic stochastic optimization problem of a nonstandard type, whose optimal solution features active learning. The proof of optimality and the derivation of the corresponding control policies is an indirect one, which relates the original single-person optimization problem to a sequence of nested zero-sum stochastic games. Existence of saddle points for these games implies the existence of optimal policies for the original control problem, which, in turn, can be obtained from the solution of a nonlinear deterministic, optimal control problem. The author also studies the problem of existence of stationary optimal policies when the time horizon is infinite and the objective function is discounted. >

Language learning as problem solving: modelling logical aspects of inductive learning to generate sentences in French by man and machine

Abstract: We present here a system under development, the present goals of which are to assist (a) students in inductively learning a set of rules to generate sentences in French, and (b) psychologists in gathering data on natural language learning.Instead of claiming an all-encompassing model or theory, we prefer to elaborate a tool, which is general and flexible enough to permit the testing of various theories. By controlling parameters such as initial knowledge, the nature and order of the data, we can empirically determine how each parameter affects the efficiency of learning. Our ultimate goal is the modelling of human learning by machine.Learning is viewed as problem-solving, i.e. as the creation and reduction of a search-space. By integrating the student into the process, that is, by encouraging him to ask an expert (the system) certain kinds of questions, like: can one say x ? how does one say x ? why does one say x ?, we can enhance not only the efficiency of the learning, but also our understanding of the underlying processes. By having a trace of the whole dialogue (what questions have been asked at what time), we should be able to infer the student's learning strategies.

Towards intelligent machine learning algorithms

Abstract: : Machine learning is recognized as a tool for improving the performance of many kinds of systems, yet most machine learning systems themselves are not well equipped to improve their own learning performance. By emphasizing the role of domain knowledge, learning systems can be crafted as knowledge directed systems, and with the addition of a knowledge store for organizing and maintaining knowledge to assist learning, a learning machine learning (L-ML) algorithm is possible. The necessary components of L-ML systems are presented along with several case descriptions of existing machine learning systems that possess limited L-ML capabilities. Keywords: Algorithms, Artificial intelligence. (SDW)

Constructive learning with continuous-values attributes

Abstract: In this paper we will present a methodology for dealing with continuous-valued attributes in constructive Concept Learning. This technique allows the system to use partially defined predicates, and set automatically the values of their parameters. In this way the user can define more easily the concept description language, and the search for discriminant expressions can be more effective. The method we propose is also a way of integrating statistical and symbolic approaches to Machine Learning, by using a description language based on first order logic, and an induction method which is also able to deal with numerical data. Although many related research issues still need to be investigated, this technique can be useful, and an example of a Concept Acquisition problem is given, where the proposed methodology is important, in order to obtain an acceptable solution.

Detecting abnormal situations from real time power plant data using machine learning

Abstract: Power plant data is recorded through hundreds of channels at a very fast rate. Many times situations arise where faulty data gets recorded. The possible causes for these situations include faulty instruments, malfunctioning plant components , faulty sensors, or operator induced errors. Identifying and removing the faulty instances from a very large volume of data is a complex task which cannot be performed manually by domain experts. Even precise characterization of these situations is difficult for a domain expert because of the large number of possible variations and incomplete knowledge about these variations. For the same reasons, statistical reduction and smoothing methods of data analysis do not detect these abnormalities. We are currently developing an expert system which learns the normal and abnormal characteristics from given examples and employs this knowledge to detect, analyze and record abnormal situations from real-time data collected for the MHD Coal Fired Flow facility at UTSI. The research work reported in this Permission to copy without fee all or part of this material is granted provided that the copies are not made or distributed for direct commercial advantage, the ACM copyright notice and the title of the publication and its date appear, and notice is given that copying is by permission of the Association for Computing Machinery. To copy otherwise, or to republish, requires a fee and/or specfic permission. paper deals with the conceptualized characterization of the faulty patterns which have close associations with the abnormal and faulty situations. Initially domain experts broadly characterize the types of patterns which may exist in the normal and abnormal situations. The expert system transforms these broad characteristics into specific conceptual descriptions which are then generalized as well as specialized in order to represent examples of the abnormal and normal situations. The output of this processing is produced in the form of characteristic and discriminant descriptors. The characteristic descriptors characterize all the examples of the situations which domain experts designate as abnormal situations. The discriminant descriptors differentiate the examples of normal and abnormal situations. The expert system employs these descriptors to scan and analyze real time data. The abnormal situations are identified and separated. If any segment of the data does not satisfy the known descriptors, domain experts are asked to characterize these new situations. The interactive response results in learning new descrip-tors.

Learning from a plan-based interface

Abstract: The use of plan recognition in the user interface is presented as a basis for a learning environment within which students can assimilate the range of possible actions and objectives afforded by a target application. Support in the form of “present whereabouts” and “possible progressions” enable students to learn by doing. When combined with a facility for exploratory learning, this helps avoid the onerous air of a “teaching situation”.

Multinomial conjunctoid statistical learning machines

Abstract: Multinomial Conjunctoids are supervised statistical modules that learn the relationships among binary events. The multinomial conjunctoid algorithm precludes the following problems that occur in existing feedforward multi-layered neural networks: (a) existing networks often cannot determine underlying neural architectures, for example how many hidden layers should be used, how many neurons in each hidden layer are required, and what interconnections between neurons should be made; (b) existing networks cannot avoid convergence to suboptimal solutions during the learning process; (c) existing networks require many iterations to converge, if at all, to stable states; and (d) existing networks may not be sufficiently general to reflect all learning situations. By contrast, multinomial conjunctoids are based on a well-developed statistical decision theory framework, which guarantees that learning algorithms will converge to optimal learning states as the number of learning trials increases, and that convergence during each trial will be very fast. In this paper a prototype multinomial conjunctoid circuit based on CMOS VLSI technology is described.

Machine Learning in a Dynamic World: Panel Discussion

Abstract: Machine Learning was established as a research discipline in the 1970's and experienced a growth expansion in the 1980's. One of the roots of machine learning research was in Cybernetic Systems and Adaptive Control. Machine Learning has been significantly influenced by Artificial Intelligence, Cognitive Science, Computer Science, and other disciplines; Machine Learning has developed its own research paradigms, methodologies, and a set of research objectives different from those of control systems. In the meantime, a new field of Intelligent Control has emerged. Even though Intelligent Control adheres more closely to the traditional control systems theory paradigms - mainly quantitative descriptions, differential equations models, goal-oriented system design, rigid mathematical formulation of goals and models - it has also deviated from the traditional systems theory approach. The two fields have moved forward without much interaction between them - different conferences, different journals, different researchers. Machine Learning has been concerned primarily with general learning mechanisms and methodologies for software implementation of learning systems. Intelligent Control has concentrated more on the dynamics of real physical systems and practical control problem solving. Because the two disciplines have at least me goal in common - automatic acquisition of knowledge about the world - they should have more interaction. The lack of interdisciplinary communication may lead to some undesirable results: establishing different terminologies for the same phenomena, repetitive work (discovering the same things independently), and lower quality research (ignoring the results established by the other discipline). The goal of this panel was to analyze the interactions between Machine Learning and Intelligent Control. The panel consisted of several researchers both from the area of Intelligent Control and from Machine Learning. The panelists were asked to concentrate on such general issues

A notion for machine learning: knowledge developability

Abstract: A notion, called knowledge development, for measuring the intelligence of machines is studied in the context of designing a versatile learning system. The notion is used in several ways: (1) it provides a guide to verify the overall intelligence of the expected machine, (2) it is claimed that an intelligently learning system should be able to assess the gain of intelligence level when a piece of knowledge is to be learned, and the notion can be used to address the feasibility of such an assessment and (3) it draws a justifiable boundary between the machine's responsibility and the user's responsibility. The base model of a general computing machine used is a Mealy finite state machine. >

Machine Learning Techniques for Knowledge-Based Systems.

Abstract: This paper provides an overview of recent research in developing machine learning techniques that are relevant to creating knowledge-based systems.