Function Approximation In Reinforcement Learning Research Articles

Scalable Scheduling of Semiconductor Packaging Facilities Using Deep Reinforcement Learning.

Reinforcement learning (RL) has emerged as a promising approach for scheduling semiconductor operations. Yet, it is still challenging to solve large-scale scheduling problems based on an RL method since learning complexity grows fast as the size of shop floor increases. This challenge becomes more apparent when solving the scheduling problems with a diverse number of job types, which leads to the difficulties in exploration and function approximation in RL. This article presents a scheduling method for semiconductor packaging facilities using deep RL in which an agent allocates a job to one of machines in a centralized manner. Specifically, a novel state representation is introduced to effectively accommodate the variations in the number of available machines and the production requirements. Furthermore, we propose a continuous representation of an action to maintain the size of the action space even when the numbers of jobs, machines, and operation types are subject to change. Extensive experiments on large-scale datasets demonstrate that the proposed method mostly outperforms the metaheuristics and rule-based methods, as well as the other RL approaches considered in terms of makespan while requiring much less computation time than the metaheuristics.

Data-driven optimal control of wind turbines using reinforcement learning with function approximation

We propose a reinforcement learning approach with function approximation for maximizing the power output of wind turbines (WTs). The optimal control of wind turbines majorly uses the maximum power point tracking (MPPT) strategy for sequential decision-making that can be modeled as a Markov decision process (MDP). In the literature, the continuous control variables are typically discretized to cope with the curse of dimensionality in traditional dynamic programming methods. To provide a more accurate prediction, we formulate the problem into an MDP with continuous state and action spaces by utilizing the function approximation in reinforcement learning. The commonly used pitch angle is selected as a control variable we are concerned with, which is regarded as the system state along with some other controllable and uncontrollable variables proven to affect the power output. Computational studies of real data are conducted to demonstrate that the proposed method outperforms the existing methods in the literature in obtaining the optimal power output.

PoPS: Policy Pruning and Shrinking for Deep Reinforcement Learning

The recent success of deep neural networks (DNNs) for function approximation in reinforcement learning has triggered the development of Deep Reinforcement Learning (DRL) algorithms in various fields, such as robotics, computer games, natural language processing, computer vision, sensing systems, and wireless networking. Unfortunately, DNNs suffer from high computational cost and memory consumption, which limits the use of DRL algorithms in systems with limited hardware resources. In recent years, pruning algorithms have demonstrated considerable success in reducing the redundancy of DNNs in classification tasks. However, existing algorithms suffer from a significant performance reduction in the DRL domain. In this article, we develop the first effective solution to the performance reduction problem of pruning in the DRL domain, and establish a working algorithm, named Policy Pruning and Shrinking (PoPS), to train DRL models with strong performance while achieving a compact representation of the DNN. The framework is based on a novel iterative policy pruning and shrinking method that leverages the power of transfer learning when training the DRL model. We present an extensive experimental study that demonstrates the strong performance of PoPS using the popular Cartpole, Lunar Lander, Pong, and Pacman environments. Finally, we develop an open source software for the benefit of researchers and developers in related fields.

Sigmoid-weighted linear units for neural network function approximation in reinforcement learning

In recent years, neural networks have enjoyed a renaissance as function approximators in reinforcement learning. Two decades after Tesauro’s TD-Gammon achieved near top-level human performance in backgammon, the deep reinforcement learning algorithm DQN achieved human-level performance in many Atari 2600 games. The purpose of this study is twofold. First, we propose two activation functions for neural network function approximation in reinforcement learning: the sigmoid-weighted linear unit (SiLU) and its derivative function (dSiLU). The activation of the SiLU is computed by the sigmoid function multiplied by its input. Second, we suggest that the more traditional approach of using on-policy learning with eligibility traces, instead of experience replay, and softmax action selection can be competitive with DQN, without the need for a separate target network. We validate our proposed approach by, first, achieving new state-of-the-art results in both stochastic SZ-Tetris and Tetris with a small 10 × 10 board, using TD(λ) learning and shallow dSiLU network agents, and, then, by outperforming DQN in the Atari 2600 domain by using a deep Sarsa(λ) agent with SiLU and dSiLU hidden units.

From free energy to expected energy: Improving energy-based value function approximation in reinforcement learning

Free-energy based reinforcement learning (FERL) was proposed for learning in high-dimensional state and action spaces. However, the FERL method does only really work well with binary, or close to binary, state input, where the number of active states is fewer than the number of non-active states. In the FERL method, the value function is approximated by the negative free energy of a restricted Boltzmann machine (RBM). In our earlier study, we demonstrated that the performance and the robustness of the FERL method can be improved by scaling the free energy by a constant that is related to the size of network. In this study, we propose that RBM function approximation can be further improved by approximating the value function by the negative expected energy (EERL), instead of the negative free energy, as well as being able to handle continuous state input. We validate our proposed method by demonstrating that EERL: (1) outperforms FERL, as well as standard neural network and linear function approximation, for three versions of a gridworld task with high-dimensional image state input; (2) achieves new state-of-the-art results in stochastic SZ-Tetris in both model-free and model-based learning settings; and (3) significantly outperforms FERL and standard neural network function approximation for a robot navigation task with raw and noisy RGB images as state input and a large number of actions.

Manifold-Based Reinforcement Learning via Locally Linear Reconstruction.

Feature representation is critical not only for pattern recognition tasks but also for reinforcement learning (RL) methods to solve learning control problems under uncertainties. In this paper, a manifold-based RL approach using the principle of locally linear reconstruction (LLR) is proposed for Markov decision processes with large or continuous state spaces. In the proposed approach, an LLR-based feature learning scheme is developed for value function approximation in RL, where a set of smooth feature vectors is generated by preserving the local approximation properties of neighboring points in the original state space. By using the proposed feature learning scheme, an LLR-based approximate policy iteration (API) algorithm is designed for learning control problems with large or continuous state spaces. The relationship between the value approximation error of a new data point and the estimated values of its nearest neighbors is analyzed. In order to compare different feature representation and learning approaches for RL, a comprehensive simulation and experimental study was conducted on three benchmark learning control problems. It is illustrated that under a wide range of parameter settings, the LLR-based API algorithm can obtain better learning control performance than the previous API methods with different feature representation schemes.

A clustering-based graph Laplacian framework for value function approximation in reinforcement learning.

In order to deal with the sequential decision problems with large or continuous state spaces, feature representation and function approximation have been a major research topic in reinforcement learning (RL). In this paper, a clustering-based graph Laplacian framework is presented for feature representation and value function approximation (VFA) in RL. By making use of clustering-based techniques, that is, K-means clustering or fuzzy C-means clustering, a graph Laplacian is constructed by subsampling in Markov decision processes (MDPs) with continuous state spaces. The basis functions for VFA can be automatically generated from spectral analysis of the graph Laplacian. The clustering-based graph Laplacian is integrated with a class of approximation policy iteration algorithms called representation policy iteration (RPI) for RL in MDPs with continuous state spaces. Simulation and experimental results show that, compared with previous RPI methods, the proposed approach needs fewer sample points to compute an efficient set of basis functions and the learning control performance can be improved for a variety of parameter settings.

Adaptive function approximation in reinforcement learning with an interpolating growing neural gas1

Q-Learning is a widely used method for dealing with reinforcement learning problems. To speed up learning and to exploit gained experience more efficiently it is highly beneficial to add generalization to Q-Learning and thus enabling the transfer of

Feature Search in the Grassmanian in Online Reinforcement Learning

We consider the problem of finding the best features for value function approximation in reinforcement learning and develop an online algorithm to optimize the mean square Bellman error objective. For any given feature value, our algorithm performs gradient search in the parameter space via a residual gradient scheme and, on a slower timescale, also performs gradient search in the Grassman manifold of features. We present a proof of convergence of our algorithm. We show empirical results using our algorithm as well as a similar algorithm that uses temporal difference learning in place of the residual gradient scheme for the faster timescale updates.

Basis Function Discovery Using Spectral Clustering and Bisimulation Metrics

We study the problem of automatically generating features for function approximation in reinforcement learning. We build on the work of Mahadevan and his colleagues, who pioneered the use of spectral clustering methods for basis function construction. Their methods work on top of a graph that captures state adjacency. Instead, we use bisimulation metrics in order to provide state distances for spectral clustering. The advantage of these metrics is that they incorporate reward information in a natural way, in addition to the state transition information. We provide theoretical bounds on the quality of the obtained approximation, which justify the importance of incorporating reward information. We also demonstrate empirically that the approximation quality improves when bisimulation metrics are used instead of the state adjacency graph in the basis function construction process.

Efficient exploration through active learning for value function approximation in reinforcement learning

Appropriately designing sampling policies is highly important for obtaining better control policies in reinforcement learning. In this paper, we first show that the least-squares policy iteration (LSPI) framework allows us to employ statistical active learning methods for linear regression. Then we propose a design method of good sampling policies for efficient exploration, which is particularly useful when the sampling cost of immediate rewards is high. The effectiveness of the proposed method, which we call active policy iteration (API), is demonstrated through simulations with a batting robot.

Fuzzy decision tree function approximation in reinforcement learning

Recent results on reinforcement learning regarding the convergence of control algorithms with function approximators, have shown that decision tree based reinforcement learning provides good learning performance and more reliable convergence than the neural network approach. It scales better to larger input spaces with lower memory requirements, and can solve problems that are infeasible using table lookup. However, decision tree based reinforcement learning can deal with only discrete actions. In realistic applications, it is imperative to deal with continuous states and actions. In this paper, we have proposed fuzzy decision tree based reinforcement learning that takes care of the limitations of decision tree based learning. We compare our approach with decision tree based function approximator on two bench mark problems: inverted pendulum stabilisation problem and two-link robot manipulator tracking problem.

Reinforcement Learning in Continuous Time and Space: Interference and Not Ill Conditioning Is the Main Problem When Using Distributed Function Approximators

Many interesting problems in reinforcement learning (RL) are continuous and/or high dimensional, and in this instance, RL techniques require the use of function approximators for learning value functions and policies. Often, local linear models have been preferred over distributed nonlinear models for function approximation in RL. We suggest that one reason for the difficulties encountered when using distributed architectures in RL is the problem of negative interference, whereby learning of new data disrupts previously learned mappings. The continuous temporal difference (TD) learning algorithm TD(lambda) was used to learn a value function in a limited-torque pendulum swing-up task using a multilayer perceptron (MLP) network. Three different approaches were examined for learning in the MLP networks; 1) simple gradient descent; 2) vario-eta; and 3) a pseudopattern rehearsal strategy that attempts to reduce the effects of interference. Our results show that MLP networks can be used for value function approximation in this task but require long training times. We also found that vario-eta destabilized learning and resulted in a failure of the learning process to converge. Finally, we showed that the pseudopattern rehearsal strategy drastically improved the speed of learning. The results indicate that interference is a greater problem than ill conditioning for this task.