Showing papers by "Charles W. Anderson published in 1988"

Neuronlike adaptive elements that can solve difficult learning control problems

Abstract: It is shown how a system consisting of two neuronlike adaptive elements can solve a difficult learning control problem. The task is to balance a pole that is hinged to a movable cart by applying forces to the cart's base. It is argued that the learning problems faced by adaptive elements that are components of adaptive networks are at least as difficult as this version of the pole-balancing problem. The learning system consists of a single associative search element (ASE) and a single adaptive critic element (ACE). In the course of learning to balance the pole, the ASE constructs associations between input and output by searching under the influence of reinforcement feedback, and the ACE constructs a more informative evaluation function than reinforcement feedback alone can provide. The differences between this approach and other attempts to solve problems using neurolike elements are discussed, as is the relation of this work to classical and instrumental conditioning in animal learning studies and its possible implications for research in the neurosciences.

Learning to Control an Inverted Pendulum with Connectionist Networks

Abstract: An inverted pendulum is simulated and cast as a control task with the goal of learning to avoid a subset of states with no a priori knowledge of the pendulum's dynamics. To solve this task a controller must deal with the issues of delayed performance evaluation, learning under uncertainty, and the learning of nonlinear functions. These issues are addressed by connectionist learing procedures that learn to balance the pendulum.

Application of connectionist learning methods to manufacturing process monitoring

Abstract: It is demonstrated that connectionist learning networks can monitor manufacturing processes to determine causal relationships with an accuracy competitive with that of conventional statistical techniques. Moreover, the network operates online, in realtime, and with substantial savings in computational complexity as compared with conventional CIM techniques. Two approaches are compared. One employs standard procedures to find correlations between sensor measurements and quality. The sensor data from the production line are collected over a period of time, and correlations are made offline at infrequent intervals using analyses such as linear regression. The second approach is to estimate the correlations incrementally, as the data are collected, online and in real-time. The estimates are updated incrementally using connectionist learning procedures. Simulation results are presented for a fluorescent bulb manufacturing line. >