Mutation Operator Evolution for EA-Based Neural Networks
نویسنده
چکیده
In the field of artificial intelligence, one of the hardest things that we can try to make a computer program do is to interact with the real world. In contrast to the well-defined, discrete, simplified world that programs normally operate in, the real world is large, unknown, and complex. In the real world, programs must learn and adapt to new and changing situations in order to be effective. Reinforcement Learning is a set of methods that attempts to bridge this gap between computers and the real world. In Reinforcement learning a program, or agent, acquires information about the state of the environment and chooses an action based on that state. The environment responds to the action by returning a reward, which may be positive or negative, to the agent. The agent then uses the reward to modify what action it takes the next time it is in a similar state. Over infinite time, the agent can learns an optimal policy of action for each state in the environment[SuttonBarto1998](Pg.52-85). Many techniques exist for implementing reinforcement learning, and most center around the construction of a state-value function that records the value of given state. Given a state, the agent then uses the value function to determine the action from that state that will result in the highest reward. In situations where the environment can be represented discretely, or is small and finite, a simple table can represent the value function for every state in the environment. However, many real-world environments are neither discrete or finite, and alternate methods for finding the value function must be used. Since the environment is continuous, and possibly infinite, it is possible that in a finite amount of time the value function may never be exactly determined. Thus, we must turn to function approximation techniques for practical reinforcement learning implementations. One of the most popular function approximation methods is that of Artificial Neural Networks[SuttonBarto1998](Pg.193226). Traditionally, ANN’s have been used in conjunction with gradient-decent methods and error-backpropagation. These techniques both require knowledge of the function to be approximated in order to be successful. Since the exact characteristics of the value function may not be known beforehand, these errorbased techniques are inadequate. Also, for problem environments that change
منابع مشابه
A Differential Evolution and Spatial Distribution based Local Search for Training Fuzzy Wavelet Neural Network
Abstract Many parameter-tuning algorithms have been proposed for training Fuzzy Wavelet Neural Networks (FWNNs). Absence of appropriate structure, convergence to local optima and low speed in learning algorithms are deficiencies of FWNNs in previous studies. In this paper, a Memetic Algorithm (MA) is introduced to train FWNN for addressing aforementioned learning lacks. Differential Evolution...
متن کاملA New Self-Adaptive EP Approach for ANN Weights Training
Evolutionary Programming (EP) represents a methodology of Evolutionary Algorithms (EA) in which mutation is considered as a main reproduction operator. This paper presents a novel EP approach for Artificial Neural Networks (ANN) learning. The proposed strategy consists of two components: the self-adaptive, which contains phenotype information and the dynamic, which is described by genotype. Sel...
متن کاملImage Backlight Compensation Using Recurrent Functional Neural Fuzzy Networks Based on Modified Differential Evolution
In this study, an image backlight compensation method using adaptive luminance modification is proposed for efficiently obtaining clear images.The proposed method combines the fuzzy C-means clustering method, a recurrent functional neural fuzzy network (RFNFN), and a modified differential evolution.The proposed RFNFN is based on the two backlight factors that can accurately detect the compensat...
متن کاملDesign of Neural Networks by the Extended Differential Evolution Algorithm
This paper proposes an improved evolutionary method to optimize the beta function neural network (BBFNN) in order to approach target function from a set of input-output. The procedure allows the application of the Improved Differential Evolution (IDE) as the optimization engine of our algorithm to improve the solution of the problem at hand by including the extended mutation and crossover opera...
متن کاملNavigation of a Mobile Robot Using Virtual Potential Field and Artificial Neural Network
Mobile robot navigation is one of the basic problems in robotics. In this paper, a new approach is proposed for autonomous mobile robot navigation in an unknown environment. The proposed approach is based on learning virtual parallel paths that propel the mobile robot toward the track using a multi-layer, feed-forward neural network. For training, a human operator navigates the mobile robot in ...
متن کامل