Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “learned”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,369 records · Page 76

Learning in tele-autonomous systems using Soar

Robo-Soar is a high-level robot arm control system implemented in Soar. Robo-Soar learns to perform simple block manipulation tasks using advice from a human. Following learning, the system is able to perform similar tasks without external guidance. It can also learn to correct its knowledge, using its own problem solving in addition to outside guidance. Robo-Soar corrects its knowledge by accepting advice about relevance of features in its domain, using a unique integration of analytic and empirical learning techniques.

Laird, John E.↗

Neural networks for self-learning control systems

It is shown how a neural network can learn of its own accord to control a nonlinear dynamic system. An emulator, a multilayered neural network, learns to identify the system's dynamic characteristics. The controller, another multilayered neural network, next learns to control the emulator. The self-trained controller is then used to control the actual dynamic system. The learning process continues as the emulator and controller improve and track the physical process. An example is given to illustrate these ideas. The 'truck backer-upper,' a neural network controller that steers a trailer truck while the truck is backing up to a loading dock, is demonstrated. The controller is able to guide the truck to the dock from almost any initial position. The technique explored should be applicable to a wide variety of nonlinear control problems.

Nguyen, Derrick H.↗

Learning from external environments using Soar

Soar, like the previous PRODIGY and Theo, is a problem-solving architecture that attempts to learn from experience; unlike them, it takes a more uniform approach, using a single forward-chaining architecture for planning and execution. Its single learning mechanism, designated 'chunking', is domain-independent. Two developmental approaches have been employed with Soar: the first of these allows the architecture to attempt a problem on its own, while the second involves a degree of external guidance. This learning through guidance is integrated with general problem-solving and autonomous learning, leading to an avoidance of human interaction for simple problems that Soar can solve on its own.

Laird, John E.↗

Learning control for minimizing a quadratic cost during repetitions of a task

In many applications, control systems are asked to perform the same task repeatedly. Learning control laws have been developed over the last few years that allow the controller to improve its performance each repetition, and to converge to zero error in tracking a desired trajectory. This paper generates a new type of learning control law that learns to minimize a quadratic cost function for tracking. Besides being of interest in its own right, this objective alleviates the need to specify a desired trajectory that can actually be performed by the system. The approach used here is to adapt appropriate methods from numerical optimization theory in order to produce learning control algorithms that adjust the system command from repetition to repetition in order to converge to the quadratic cost optimal trajectory.

Longman, Richard W.↗

Accelerating Learning By Neural Networks

Electronic neural networks made to learn faster by use of terminal teacher forcing. Method of supervised learning involves addition of teacher forcing functions to excitations fed as inputs to output neurons. Initially, teacher forcing functions are strong enough to force outputs to desired values; subsequently, these functions decay with time. When learning successfully completed, terminal teacher forcing vanishes, and dynamics or neural network become equivalent to those of conventional neural network. Simulated neural network with terminal teacher forcing learned to produce close approximation of circular trajectory in 400 iterations.

Toomarian, Nikzad↗

Paradigms for machine learning

Five paradigms are described for machine learning: connectionist (neural network) methods, genetic algorithms and classifier systems, empirical methods for inducing rules and decision trees, analytic learning methods, and case-based approaches. Some dimensions are considered along with these paradigms vary in their approach to learning, and the basic methods are reviewed that are used within each framework, together with open research issues. It is argued that the similarities among the paradigms are more important than their differences, and that future work should attempt to bridge the existing boundaries. Finally, some recent developments in the field of machine learning are discussed, and their impact on both research and applications is examined.

Schlimmer, Jeffrey C.↗

Learning class descriptions from a data base of spectral reflectance with multiple view angles

A learning program has been developed which combines 'learning by example' with the generate-and-test paradigm to furnish a robust learning environment capable of handling error-prone data. The problem is shown to be capable of learning class descriptions from positive and negative training examples of spectral and directional reflectance data taken from soil and vegetation. The program, which used AI techniques to automate very tedious processes, found the sequence of relationships that contained the most important information which could distinguish the classes.

Kimes, Daniel S.↗

Techniques for improving transients in learning control systems

A discrete modern control formulation is used to study the nature of the transient behavior of the learning process during repetitions. Several alternative learning control schemes are developed to improve the transient performance. These include a new method using an alternating sign on the learning gain, which is very effective in limiting peak transients and also very useful in multiple-input, multiple-output systems. Other methods include learning at an increasing number of points progressing with time, or an increasing number of points of increasing density.

Chang, C.-K.↗

Discrete time learning control in nonlinear systems

In this paper digital learning control methods are developed primarily for use in single-input, single-output nonlinear dynamic systems. Conditions for convergence of the basic form of learning control based on integral control concepts are given, and shown to be satisfied by a large class of nonlinear problems. It is shown that it is not the gross nonlinearities of the differential equations that matter in the convergence, but rather the much smaller nonlinearities that can manifest themselves during the short time interval of one sample time. New algorithms are developed that eliminate restrictions on the size of the learning gain, and on knowledge of the appropriate sign of the learning gain, for convergence to zero error in tracking a feasible desired output trajectory. It is shown that one of the new algorithms can give guaranteed convergence in the presence of actuator saturation constraints, and indicate when the requested trajectory is beyond the actuator capabilities.

Longman, Richard W.↗

Learning to improve iterative repair scheduling

This paper presents a general learning method for dynamically selecting between repair heuristics in an iterative repair scheduling system. The system employs a version of explanation-based learning called Plausible Explanation-Based Learning (PEBL) that uses multiple examples to confirm conjectured explanations. The basic approach is to conjecture contradictions between a heuristic and statistics that measure the quality of the heuristic. When these contradictions are confirmed, a different heuristic is selected. To motivate the utility of this approach we present an empirical evaluation of the performance of a scheduling system with respect to two different repair strategies. We show that the scheduler that learns to choose between the heuristics outperforms the same scheduler with any one of two heuristics alone.

Zweben, Monte↗

Learning fuzzy information in a hybrid connectionist, symbolic model

An instance-based learning system is presented. SC-net is a fuzzy hybrid connectionist, symbolic learning system. It remembers some examples and makes groups of examples into exemplars. All real-valued attributes are represented as fuzzy sets. The network representation and learning method is described. To illustrate this approach to learning in fuzzy domains, an example of segmenting magnetic resonance images of the brain is discussed. Clearly, the boundaries between human tissues are ill-defined or fuzzy. Example fuzzy rules for recognition are generated. Segmentations are presented that provide results that radiologists find useful.

Romaniuk, Steve G.↗

Design issues for a reinforcement-based self-learning fuzzy controller

Fuzzy logic controllers have some often cited advantages over conventional techniques such as PID control: easy implementation, its accommodation to natural language, the ability to cover wider range of operating conditions and others. One major obstacle that hinders its broader application is the lack of a systematic way to develop and modify its rules and as result the creation and modification of fuzzy rules often depends on try-error or pure experimentation. One of the proposed approaches to address this issue is self-learning fuzzy logic controllers (SFLC) that use reinforcement learning techniques to learn the desirability of states and to adjust the consequent part of fuzzy control rules accordingly. Due to the different dynamics of the controlled processes, the performance of self-learning fuzzy controller is highly contingent on the design. The design issue has not received sufficient attention. The issues related to the design of a SFLC for the application to chemical process are discussed and its performance is compared with that of PID and self-tuning fuzzy logic controller.

Yen, John↗

Using neural networks and Dyna algorithm for integrated planning, reacting and learning in systems

The traditional AI answer to the decision making problem for a robot is planning. However, planning is usually CPU-time consuming, depending on the availability and accuracy of a world model. The Dyna system generally described in earlier work, uses trial and error to learn a world model which is simultaneously used to plan reactions resulting in optimal action sequences. It is an attempt to integrate planning, reactive, and learning systems. The architecture of Dyna is presented. The different blocks are described. There are three main components of the system. The first is the world model used by the robot for internal world representation. The input of the world model is the current state and the action taken in the current state. The output is the corresponding reward and resulting state. The second module in the system is the policy. The policy observes the current state and outputs the action to be executed by the robot. At the beginning of program execution, the policy is stochastic and through learning progressively becomes deterministic. The policy decides upon an action according to the output of an evaluation function, which is the third module of the system. The evaluation function takes the following as input: the current state of the system, the action taken in that state, the resulting state, and a reward generated by the world which is proportional to the current distance from the goal state. Originally, the work proposed was as follows: (1) to implement a simple 2-D world where a 'robot' is navigating around obstacles, to learn the path to a goal, by using lookup tables; (2) to substitute the world model and Q estimate function Q by neural networks; and (3) to apply the algorithm to a more complex world where the use of a neural network would be fully justified. In this paper, the system design and achieved results will be described. First we implement the world model with a neural network and leave Q implemented as a look up table. Next, we use a lookup table for the world model and implement the Q function with a neural net. Time limitations prevented the combination of these two approaches. The final section discusses the results and gives clues for future work.

Lima, Pedro↗

Design issues of a reinforcement-based self-learning fuzzy controller for petrochemical process control

Fuzzy logic controllers have some often-cited advantages over conventional techniques such as PID control, including easier implementation, accommodation to natural language, and the ability to cover a wider range of operating conditions. One major obstacle that hinders the broader application of fuzzy logic controllers is the lack of a systematic way to develop and modify their rules; as a result the creation and modification of fuzzy rules often depends on trial and error or pure experimentation. One of the proposed approaches to address this issue is a self-learning fuzzy logic controller (SFLC) that uses reinforcement learning techniques to learn the desirability of states and to adjust the consequent part of its fuzzy control rules accordingly. Due to the different dynamics of the controlled processes, the performance of a self-learning fuzzy controller is highly contingent on its design. The design issue has not received sufficient attention. The issues related to the design of a SFLC for application to a petrochemical process are discussed, and its performance is compared with that of a PID and a self-tuning fuzzy logic controller.

Yen, John↗

Learning class descriptions from a data base of spectral reflectance of soil samples

Consideration is given to a program developed to learn class descriptions from positive and negative training examples of spectral reflectance data of bare soils. It is a combination of 'learning by example' and the generate-and-test paradigm and is designed to provide a robust learning environment that can handle error-prone data. The program was tested by having it learn class descriptions of various categories of organic carbon content, iron oxide content, and particle size distribution in soils. These class descriptions were then used to classify an array of targets. The program found the sequence of relationships between bands that contained the most important information to distinguish the classes. Physical explanations for the class descriptions obtained are presented.

Kimes, D. S.↗

On the asymptotic improvement of supervised learning by utilizing additional unlabeled samples - Normal mixture density case

The effect of additional unlabeled samples in improving the supervised learning process is studied in this paper. Three learning processes. supervised, unsupervised, and combined supervised-unsupervised, are compared by studying the asymptotic behavior of the estimates obtained under each process. Upper and lower bounds on the asymptotic covariance matrices are derived. It is shown that under a normal mixture density assumption for the probability density function of the feature space, the combined supervised-unsupervised learning is always superior to the supervised learning in achieving better estimates. Experimental results are provided to verify the theoretical concepts.

Shahshahani, Behzad M.↗

Learning and optimization with cascaded VLSI neural network building-block chips

To demonstrate the versatility of the building-block approach, two neural network applications were implemented on cascaded analog VLSI chips. Weights were implemented using 7-b multiplying digital-to-analog converter (MDAC) synapse circuits, with 31 x 32 and 32 x 32 synapses per chip. A novel learning algorithm compatible with analog VLSI was applied to the two-input parity problem. The algorithm combines dynamically evolving architecture with limited gradient-descent backpropagation for efficient and versatile supervised learning. To implement the learning algorithm in hardware, synapse circuits were paralleled for additional quantization levels. The hardware-in-the-loop learning system allocated 2-5 hidden neurons for parity problems. Also, a 7 x 7 assignment problem was mapped onto a cascaded 64-neuron fully connected feedback network. In 100 randomly selected problems, the network found optimal or good solutions in most cases, with settling times in the range of 7-100 microseconds.

Duong, T.↗

A new learning strategy for the two-time-scale neural controller with its application to the tracking control of rigid arms

A novel fast learning rule with fast weight identification is proposed for the two-time-scale neural controller, and a two-stage learning strategy is developed for the proposed neural controller. The results of the stability analysis show that both the tracking error and the fast weight error will be uniformly bounded and converge to a bounded region which depends only on the accuracy of the slow learning if the system is sufficiently excited. The efficiency of the two-stage learning is also demonstrated by a simulation of a two-link arm.

Cheng, W.↗