Intelligent Parameter Tuning Using Segmented Adaptive Reinforcement Learning Algorithm

Now-a-days, intelligent systems play a crucial role in enhancing the capabilities of traditional processing techniques to handle huge amount of data and complex problems due to advances in technologies. Most of the processing problems can be resolved using traditional solutions based on optimization algorithms. In general, these solutions are normalized to increase their applications. Many of these solutions have control parameters to optimize the performance of the solution and also to maintain relative importance to specific application. Parameter tuning is a straightforward manual by determining the direction of tuning by trial and error method. But manual adjustment for these control parameters is tedious and consumes too much time and effort and sometimes becomes impractical if the solution has more parameters and requires precise tuning. With solution space being complex, machine learning has employed as a key component to setting up correct and precise parameters. Reinforcement Machine Learning algorithms can be used to solve this problem but existing algorithms requires huge amount of learning time and resources. This paper aims at solving this problem and proposes novel Segmented and Adaptive Reinforcement Learning (SARL) algorithm to train the system that can automatically tune the parameters accurately and precisely with very minimal learning time. Performance of proposed algorithm is validated in wavelet-based noise reduction technique by employing SARL algorithm to adjust 3 control parameters of Noise based Hybrid Threshold method. After integrating with the proposed SARL algorithm, the learning time of noise reduction technique is highly reduced without compromising the performance of the considered technique.