Published Paper Details:

Machine Learning algorithms play very crucial role in decision automation process of Artificial Intelligent systems. Machine Learning algorithms learn from the hidden structures present inside the data without the need of any traditional programming. Many advanced algorithms are developed for Machine Learning to handle complex datasets. These algorithms provide better performance but require huge amount of training time. This requirement of huge training time is manageable in Supervised and Unsupervised Machine Learning algorithms as their models are trained before deploying them into application. But, in Reinforcement learning, the effect of huge training time is very significant problem as their models are trained after deploying them into application environment. To solve this problem, new Reinforcement algorithms with controlled complexity need to be developed without compromising the performance of the model. This paper aims at four new Reinforcement Learning algorithms with controlled complexity to reduce the training time. The proposing algorithms are developed using MATLAB software and validated by employing them for automated parameter tuning in image denoising technique using Double Density Dual-tree Discrete Wavelet Transform. These proposing algorithms are compared against standard Markov Decision Process based Reinforcement Algorithm in terms of model accuracy and model training times