A reinforcement learning algorithm with evolving fuzzy neural networks
TL;DR: A dynamic evolving fuzzy neural network (DENFIS) function approximation approach to RL systems and results have demonstrated that the proposed approach performs well in reinforcement learning problems.
About: This article is published in IFAC Proceedings Volumes.The article was published on 2014-01-01. It has received 8 citations till now. The article focuses on the topics: Learning classifier system & Unsupervised learning.
Citations
More filters
TL;DR: In this paper, the authors enhance the notion of Dombi aggregation operators by introducing the DAOs in the interval-valued T-spherical fuzzy (IVTSF) environment where the uncertain and ambiguous information is described with the help of membership grade (MG), abstinence grade (AG), non-membership grade (NMG), and refusal grade (RG) using closed sub-intervals of [0, 1].
Abstract: Multi-attribute decision-making (MADM) is commonly used to investigate fuzzy information effectively. However, selecting the best alternative information is not always symmetric because the alternatives do not have complete information, so asymmetric information is often involved. Expressing the information under uncertainty using closed subintervals of [0, 1] is beneficial and effective instead of using crisp numbers from [0, 1]. The goal of this paper is to enhance the notion of Dombi aggregation operators (DAOs) by introducing the DAOs in the interval-valued T-spherical fuzzy (IVTSF) environment where the uncertain and ambiguous information is described with the help of membership grade (MG), abstinence grade (AG), non-membership grade (NMG), and refusal grade (RG) using closed sub-intervals of [0, 1]. One of the key benefits of the proposed work is that in the environment of information loss is reduced to a negligible limit. We proposed concepts of IVTSF Dombi weighted averaging (IVTSFDWA) and IVTSF Dombi weighted geometric (IVTSFDWG) operators. The diversity of the IVTSF DAOs is proved and the influences of the parameters, associated with DAOs, on the ranking results are observed in a MADM problem where it is discussed how a decision can be made when there is asymmetric information about alternatives.
16 citations
01 Jan 2019
TL;DR: This paper presents a concise review on implementing reinforcement learning with evolutionary algorithms and proposes a Q-value-based GRL for fuzzy controller (QGRF) where evolution is performed after each trial in contrast to GA where many trials are required to be performed before evolution.
Abstract: Evolutionary algorithms have come to take a centre stage in diverse areas spanning multiple applications. Reinforcement learning is a novel paradigm that has recently evolved as a major control technique. This paper presents a concise review on implementing reinforcement learning with evolutionary algorithms, e.g. genetic algorithm (GA), particle swarm optimization (PSO), ant colony optimization (ACO), to several benchmark control problems, e.g. inverted pendulum, cart–pole problem, mobile robots. Some techniques have combined Q-Learning with evolutionary approaches to improve their performance. Others have used knowledge acquisition to obtain optimal fuzzy rule set and genetic reinforcement learning (GRL) for designing consequent parts of fuzzy systems. We also propose a Q-value-based GRL for fuzzy controller (QGRF) where evolution is performed after each trial in contrast to GA where many trials are required to be performed before evolution.
11 citations
Journal Article•
TL;DR: This paper describes the state of the art of NNRL algorithms, with a focus on robotics applications and a comprehensive survey is started with a discussion on the concepts of RL.
Abstract: In recent years, researches on reinforcement learning (RL) have focused on bridging the gap between adaptive optimal control and bio-inspired learning techniques. Neural network reinforcement learning (NNRL) is among the most popular algorithms in the RL framework. The advantage of using neural networks enables the RL to search for optimal policies more efficiently in several real-life applications. Although many surveys investigated general RL, no survey is specifically dedicated to the combination of artificial neural networks and RL. This paper therefore describes the state of the art of NNRL algorithms, with a focus on robotics applications. In this paper, a comprehensive survey is started with a discussion on the concepts of RL. Then, a review of several different NNRL algorithms is presented. Afterwards, the performances of different NNRL algorithms are evaluated and compared in learning prediction and learning control tasks from an empirical aspect and the paper concludes with a discussion on open issues.
10 citations
TL;DR: In this paper, a soft computing technique based on fuzzy logic is used to overcome the limitations of conventional regression models for multi-objective optimization in manufacturing processes, and more accurate results than those obtained from regression techniques are obtained.
Abstract: In manufacturing engineering, it is common to use both symmetrical and asymmetrical factorial designs along with regression techniques to model technological response variables, since the in-advance prediction of their behavior is of great importance to determine the levels of variation that lead to optimal response values to be obtained. For this purpose, regression techniques based on the response surface method combined with a desirability function for multi-objective optimization are commonly employed, since it is usual to find manufacturing processes that require simultaneous optimization of several variables, which exhibit in many cases an opposite behavior. However, these regression models are sometimes not accurate enough to predict the behavior of these response variables, especially when they have significant non-linearities. To deal with this drawback, soft computing techniques are very effective in overcoming the limitations of conventional regression models. This present study is focused on the employment of a symmetrical design of experiments along with a new desirability function, which is proposed in this study, and with soft computing techniques based on fuzzy logic. It will be shown that more accurate results than those obtained from regression techniques are obtained. Moreover, this new desirability function is analyzed in this study.
3 citations
References
More filters
TL;DR: It is rigorously established that standard multilayer feedforward networks with as few as one hidden layer using arbitrary squashing functions are capable of approximating any Borel measurable function from one finite dimensional space to another to any desired degree of accuracy, provided sufficiently many hidden units are available.
18,794 citations
"A reinforcement learning algorithm ..." refers background in this paper
...Theoretical investigations have revealed that neural networks and fuzzy inference systems are universal approximators [5, 6]....
[...]
TL;DR: An additive fuzzy system can uniformly approximate any real continuous function on a compact domain to any degree of accuracy.
Abstract: An additive fuzzy system can uniformly approximate any real continuous function on a compact domain to any degree of accuracy. An additive fuzzy system approximates the function by covering its graph with fuzzy patches in the input-output state space and averaging patches that overlap. The fuzzy system computes a conditional expectation E|Y|X| if we view the fuzzy sets as random sets. Each fuzzy rule defines a fuzzy patch and connects commonsense knowledge with state-space geometry. Neural or statistical clustering systems can approximate the unknown fuzzy patches from training data. These adaptive fuzzy systems approximate a function at two levels. At the local level the neural system approximates and tunes the fuzzy rules. At the global level the rules or patches approximate the function. >
1,282 citations
"A reinforcement learning algorithm ..." refers background in this paper
...Theoretical investigations have revealed that neural networks and fuzzy inference systems are universal approximators [5, 6]....
[...]
TL;DR: It is demonstrated that DENFIS can effectively learn complex temporal sequences in an adaptive way and outperform some well-known, existing models.
Abstract: This paper introduces a new type of fuzzy inference systems, denoted as dynamic evolving neural-fuzzy inference system (DENFIS), for adaptive online and offline learning, and their application for dynamic time series prediction. DENFIS evolve through incremental, hybrid (supervised/unsupervised), learning, and accommodate new input data, including new features, new classes, etc., through local element tuning. New fuzzy rules are created and updated during the operation of the system. At each time moment, the output of DENFIS is calculated through a fuzzy inference system based on m-most activated fuzzy rules which are dynamically chosen from a fuzzy rule set. Two approaches are proposed: (1) dynamic creation of a first-order Takagi-Sugeno-type fuzzy rule set for a DENFIS online model; and (2) creation of a first-order Takagi-Sugeno-type fuzzy rule set, or an expanded high-order one, for a DENFIS offline model. A set of fuzzy rules can be inserted into DENFIS before or during its learning process. Fuzzy rules can also be extracted during or after the learning process. An evolving clustering method (ECM), which is employed in both online and offline DENFIS models, is also introduced. It is demonstrated that DENFIS can effectively learn complex temporal sequences in an adaptive way and outperform some well-known, existing models.
1,239 citations
"A reinforcement learning algorithm ..." refers background in this paper
...Dynamic evolving neural fuzzy inference system (dmEFuNN/DENFIS) [14] is a modified version of the EFuNN with the idea that, depending on the position of the input vector in the input space, a FIS for calculating the output is formed dynamically bases on m fuzzy rules that had been created during…...
[...]
...Let the row vector of matrix is denoted as and the element of is denoted as ....
[...]
08 Mar 1992
TL;DR: The Stone-Weierstrass theorem is used to prove that fuzzy systems with product inference, centroid defuzzification, and a Gaussian membership function are capable of approximating any real continuous function on a compact set to arbitrary accuracy.
Abstract: The author proves that fuzzy systems are universal approximators. The Stone-Weierstrass theorem is used to prove that fuzzy systems with product inference, centroid defuzzification, and a Gaussian membership function are capable of approximating any real continuous function on a compact set to arbitrary accuracy. This result can be viewed as an existence theorem of an optimal fuzzy system for a wide variety of problems. >
1,075 citations
Book•
01 Jan 1996TL;DR: This text is the first to combine the study of neural networks and fuzzy systems, their basics and their use, along with symbolic AI methods to build comprehensive artificial intelligence systems.
Abstract: From the Publisher:
"Covering the latest issues and achievements, this well documented, precisely presented text is timely and suitable for graduate and upper undergraduate students in knowledge engineering, intelligent systems, AI, neural networks, fuzzy systems, and related areas. The author's goal is to explain the principles of neural networks and fuzzy systems and to demonstrate how they can be applied to building knowledge-based systems for problem solving. Especially useful are the comparisons between different techniques (AI rule-based methods, fuzzy methods, connectionist methods, hybrid systems) used to solve the same or similar problems."
-- Anca Ralescu, Associate Professor of Computer Science, University of Cincinnati Neural networks and fuzzy systems are different approaches to introducing human-like reasoning into expert systems. This text is the first to combine the study of these two subjects, their basics and their use, along with symbolic AI methods to build comprehensive artificial intelligence systems.
In a clear and accessible style, Kasabov describes rule- based and connectionist techniques and then their combinations, with fuzzy logic included, showing the application of the different techniques to a set of simple prototype problems, which makes comparisons possible. A particularly strong feature of the text is that it is filled with applications in engineering, business, and finance. AI problems that cover most of the application-oriented research in the field (pattern recognition, speech and image processing, classification, planning, optimization, prediction, control, decision making, and game simulations) are discussed and illustrated with concrete examples.
Intended both as a text for advanced undergraduate and postgraduate students as well as a reference for researchers in the field of knowledge engineering, Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering has chapters structured for various levels of teaching and includes original work by the author along with the classic material.
Data sets for the examples in the book as well as an integrated software environment that can be used to solve the problems and do the exercises at the end of each chapter are available free through anonymous ftp.
977 citations