Results 11  20
of
958
Selfimproving reactive agents based on reinforcement learning, planning and teaching
 Machine Learning
, 1992
"... Abstract. To date, reinforcement learning has mostly been studied solving simple learning tasks. Reinforcement learning methods that have been studied so far typically converge slowly. The purpose of this work is thus twofold: 1) to investigate the utility of reinforcement learning in solving much ..."
Abstract

Cited by 276 (2 self)
 Add to MetaCart
Abstract. To date, reinforcement learning has mostly been studied solving simple learning tasks. Reinforcement learning methods that have been studied so far typically converge slowly. The purpose of this work is thus twofold: 1) to investigate the utility of reinforcement learning in solving much more complicated learning tasks than previously studied, and 2) to investigate methods that will speed up reinforcement learning. This paper compares eight reinforcement learning frameworks: adaptive heuristic critic (AHC) learning due to Sutton, Qlearning due to Watkins, and three extensions to both basic methods for speeding up learning. The three extensions are experience replay, learning action models for planning, and teaching. The frameworks were investigated using connectionism as an approach to generalization. To evaluate the performance of different frameworks, a dynamic environment was used as a testbed. The enviromaaent is moderately complex and nondeterministic. This paper describes these frameworks and algorithms in detail and presents empirical evaluation of the frameworks.
Generalization in Reinforcement Learning: Safely Approximating the Value Function
 Advances in Neural Information Processing Systems 7
, 1995
"... To appear in: G. Tesauro, D. S. Touretzky and T. K. Leen, eds., Advances in Neural Information Processing Systems 7, MIT Press, Cambridge MA, 1995. A straightforward approach to the curse of dimensionality in reinforcement learning and dynamic programming is to replace the lookup table with a genera ..."
Abstract

Cited by 251 (3 self)
 Add to MetaCart
To appear in: G. Tesauro, D. S. Touretzky and T. K. Leen, eds., Advances in Neural Information Processing Systems 7, MIT Press, Cambridge MA, 1995. A straightforward approach to the curse of dimensionality in reinforcement learning and dynamic programming is to replace the lookup table with a generalizing function approximator such as a neural net. Although this has been successful in the domain of backgammon, there is no guarantee of convergence. In this paper, we show that the combination of dynamic programming and function approximation is not robust, and in even very benign cases, may produce an entirely wrong policy. We then introduce GrowSupport, a new algorithm which is safe from divergence yet can still reap the benefits of successful generalization. 1 INTRODUCTION Reinforcement learningthe problem of getting an agent to learn to act from sparse, delayed rewardshas been advanced by techniques based on dynamic programming (DP). These algorithms compute a value function ...
Residual Algorithms: Reinforcement Learning with Function Approximation
 In Proceedings of the Twelfth International Conference on Machine Learning
, 1995
"... A number of reinforcement learning algorithms have been developed that are guaranteed to converge to the optimal solution when used with lookup tables. It is shown, however, that these algorithms can easily become unstable when implemented directly with a general functionapproximation system, such ..."
Abstract

Cited by 237 (5 self)
 Add to MetaCart
A number of reinforcement learning algorithms have been developed that are guaranteed to converge to the optimal solution when used with lookup tables. It is shown, however, that these algorithms can easily become unstable when implemented directly with a general functionapproximation system, such as a sigmoidal multilayer perceptron, a radialbasisfunction system, a memorybased learning system, or even a linear functionapproximation system. A new class of algorithms, residual gradient algorithms, is proposed, which perform gradient descent on the mean squared Bellman residual, guaranteeing convergence. It is shown, however, that they may learn very slowly in some cases. A larger class of algorithms, residual algorithms, is proposed that has the guaranteed convergence of the residual gradient algorithms, yet can retain the fast learning speed of direct algorithms. In fact, both direct and residual gradient algorithms are shown to be special cases of residual algorithms, and it is s...
Nearoptimal reinforcement learning in polynomial time
 Machine Learning
, 1998
"... We present new algorithms for reinforcement learning, and prove that they have polynomial bounds on the resources required to achieve nearoptimal return in general Markov decision processes. After observing that the number of actions required to approach the optimal return is lower bounded by the m ..."
Abstract

Cited by 236 (3 self)
 Add to MetaCart
We present new algorithms for reinforcement learning, and prove that they have polynomial bounds on the resources required to achieve nearoptimal return in general Markov decision processes. After observing that the number of actions required to approach the optimal return is lower bounded by the mixing time T of the optimal policy (in the undiscounted case) or by the horizon time T (in the discounted case), we then give algorithms requiring a number of actions and total computation time that are only polynomial in T and the number of states, for both the undiscounted and discounted cases. An interesting aspect of our algorithms is their explicit handling of the ExplorationExploitation tradeoff. 1
An analysis of temporaldifference learning with function approximation
 IEEE Transactions on Automatic Control
, 1997
"... We discuss the temporaldifference learning algorithm, as applied to approximating the costtogo function of an infinitehorizon discounted Markov chain. The algorithm weanalyze updates parameters of a linear function approximator online, duringasingle endless trajectory of an irreducible aperiodi ..."
Abstract

Cited by 217 (7 self)
 Add to MetaCart
We discuss the temporaldifference learning algorithm, as applied to approximating the costtogo function of an infinitehorizon discounted Markov chain. The algorithm weanalyze updates parameters of a linear function approximator online, duringasingle endless trajectory of an irreducible aperiodic Markov chain with a finite or infinite state space. We present a proof of convergence (with probability 1), a characterization of the limit of convergence, and a bound on the resulting approximation error. Furthermore, our analysis is based on a new line of reasoning that provides new intuition about the dynamics of temporaldifference learning. In addition to proving new and stronger positive results than those previously available, we identify the significance of online updating and potential hazards associated with the use of nonlinear function approximators. First, we prove that divergence may occur when updates are not based on trajectories of the Markov chain. This fact reconciles positive and negative results that have been discussed in the literature, regarding the soundness of temporaldifference learning. Second, we present anexample illustrating the possibility of divergence when temporaldifference learning is used in the presence of a nonlinear function approximator.
Stable Function Approximation in Dynamic Programming
 IN MACHINE LEARNING: PROCEEDINGS OF THE TWELFTH INTERNATIONAL CONFERENCE
, 1995
"... The success of reinforcement learning in practical problems depends on the ability tocombine function approximation with temporal difference methods such as value iteration. Experiments in this area have produced mixed results; there have been both notable successes and notable disappointments. Theo ..."
Abstract

Cited by 207 (5 self)
 Add to MetaCart
The success of reinforcement learning in practical problems depends on the ability tocombine function approximation with temporal difference methods such as value iteration. Experiments in this area have produced mixed results; there have been both notable successes and notable disappointments. Theory has been scarce, mostly due to the difficulty of reasoning about function approximators that generalize beyond the observed data. We provide a proof of convergence for a wide class of temporal difference methods involving function approximators such as knearestneighbor, and show experimentally that these methods can be useful. The proof is based on a view of function approximators as expansion or contraction mappings. In addition, we present a novel view of approximate value iteration: an approximate algorithm for one environment turns out to be an exact algorithm for a different environment.
Convergence of Stochastic Iterative Dynamic Programming Algorithms
 Neural Computation
, 1994
"... Increasing attention has recently been paid to algorithms based on dynamic programming (DP) due to the suitability of DP for learning problems involving control. In stochastic environments where the system being controlled is only incompletely known, however, a unifying theoretical account of th ..."
Abstract

Cited by 207 (8 self)
 Add to MetaCart
Increasing attention has recently been paid to algorithms based on dynamic programming (DP) due to the suitability of DP for learning problems involving control. In stochastic environments where the system being controlled is only incompletely known, however, a unifying theoretical account of the behavior of these methods has been missing. In this paper we relate DPbased learning algorithms to powerful techniques of stochastic approximation via a new convergence theorem, enabling us to establish a class of convergent algorithms to which both TD() and Qlearning belong. 1
Learning to coordinate behaviors
 In Proceedings of AAAI90
, 1990
"... We describe an algorithm which allows a behaviorbased robot to learn on the basis of positive and negative feedback when to activate its behaviors. In accordance with the philosophy of behaviorbased robots, the algorithm is completely distributed: each of the behaviors independently tries to find ..."
Abstract

Cited by 206 (3 self)
 Add to MetaCart
We describe an algorithm which allows a behaviorbased robot to learn on the basis of positive and negative feedback when to activate its behaviors. In accordance with the philosophy of behaviorbased robots, the algorithm is completely distributed: each of the behaviors independently tries to find out (i) whether it is relevant (ie. whether it is at all correlated to positive feedback) and (ii) what the conditions are under which it becomes reliable (i.e. the conditions under which it maximizes the probability of receiving positive feedback and minimizes the probability of receiving negative feedback). The algorithm has been tested successfully on an autonomous 6legged robot which had to learn how to coordinate its legs so as to walk forward. Situation of the Problem Since 1985, the MIT Mobile Robot group has advocated a radically different architecture for autonomous intelligent agents (Brooks, 1986). Instead of decomposing the architecture into functional modules, such as perception, modeling, and planning (figure 1), the architecture is decomposed into taskachieving modules, also called behaviors (figure 2). This novel approach has already demonstrated to be very successful and similar approaches have become more
Learning and Sequential Decision Making
 LEARNING AND COMPUTATIONAL NEUROSCIENCE
, 1989
"... In this report we show how the class of adaptive prediction methods that Sutton called "temporal difference," or TD, methods are related to the theory of squential decision making. TD methods have been used as "adaptive critics" in connectionist learning systems, and have been proposed as models of ..."
Abstract

Cited by 195 (10 self)
 Add to MetaCart
In this report we show how the class of adaptive prediction methods that Sutton called "temporal difference," or TD, methods are related to the theory of squential decision making. TD methods have been used as "adaptive critics" in connectionist learning systems, and have been proposed as models of animal learning in classical conditioning experiments. Here we relate TD methods to decision tasks formulated in terms of a stochastic dynamical system whose behavior unfolds over time under the influence of a decision maker's actions. Strategies are sought for selecting actions so as to maximize a measure of longterm payoff gain. Mathematically, tasks such as this can be formulated as Markovian decision problems, and numerous methods have been proposed for learning how to solve such problems. We show how a TD method can be understood as a novel synthesis of concepts from the theory of stochastic dynamic programming, which comprises the standard method for solving such tasks when a model of the dynamical system is available, and the theory of parameter estimation, which provides the appropriate context for studying learning rules in the form of equations for updating associative strengths in behavioral models, or connection weights in connectionist networks. Because this report is oriented primarily toward the nonengineer interested in animal learning, it presents tutorials on stochastic sequential decision tasks, stochastic dynamic programming, and parameter estimation.
The Helmholtz Machine
, 1995
"... Discovering the structure inherent in a set of patterns is a fundamental aim of statistical inference or learning. One fruitful approach is to build a parameterized stochastic generative model, independent draws from which are likely to produce the patterns. For all but the simplest generative model ..."
Abstract

Cited by 193 (21 self)
 Add to MetaCart
Discovering the structure inherent in a set of patterns is a fundamental aim of statistical inference or learning. One fruitful approach is to build a parameterized stochastic generative model, independent draws from which are likely to produce the patterns. For all but the simplest generative models, each pattern can be generated in exponentially many ways. It is thus intractable to adjust the parameters to maximize the probability of the observed patterns. We describe a way of finessing this combinatorial explosion by maximizing an easily computed lower bound on the probability of the observations. Our method can be viewed as a form of hierarchical selfsupervised learning that may relate to the function of bottomup and topdown cortical processing pathways.