Optimal Kernel Selection Based on GPR for Adaptive Learning of Mean Throughput Rates in LTE Networks

Machine learning models and algorithms have been employed in various applications, from prognostic scrutinizing, learning and revealing patterns in data, knowledge extracting, and knowledge deducing. One promising computationally efficient and adaptive machine learning method is the Gaussian Process Regression (GPR). An essential ingredient for tuning the GPR performance is the kernel (covariance) function. The GPR models have been widely employed in diverse regression and functional approximation purposes. However, knowing the right GPR training to examine the impacts of the kernel functions on performance during implementation remains. In order to address this problem, a stepwise approach for optimal kernel selection is presented for adaptive optimal prognostic regression learning of throughput data acquired over 4G LTE networks. The resultant learning accuracy was statistically quantified using four evaluation indexes. Results indicate that the GPR training with the mertern52 kernel function achieved the best user throughput data learning among the ten contending Kernel functions.

Brain Tumor Detection from MR Images Employing Fuzzy Graph Cut Technique

Background: This research aims at the accurate selection of the seed points from the brain MRI image for the detection of the tumor region. Since, the conventional way of manual seed selection leads to inappropriate tumor extraction therefore, fuzzy clustering technique is employed for the accurate seed selection for performing the segmentation through graph cut method. Methods: In the proposed method Fuzzy Kernel Seed Selection technique is used to define the complete brain MRI image into different groups of similar intensity. Among these groups the most accurate kernels are selected empirically that show highest resemblance with the tumor. The concept of fuzziness helps making the selection even at the boundary regions. Results: The proposed Fuzzy kernel selection technique is applied on the BraTS dataset. Among the four modalities, the proposed technique is applied on Flair images. This dataset consists of Low Grade Glioma (LGG) and High Grade Glioma (HGG) tumor images. The experiment is conducted on more than 40 images and validated by evaluating the following performance metrics: 1. Disc Similarity Coefficient (DSC), 2. Jaccard Index (JI) and 3. Positive Predictive Value (PPV). The mean DSC and PPV values obtained for LGG images are 0.89 and 0.87 respectively; and for HGG images it is 0.92 and 0.90 respectively. Conclusion: On comparing the proposed Fuzzy kernel selection graph cut technique approach with the existing techniques it is observed that the former provides an automatic accurate tumor detection. It is highly efficient and can provide a better performance for HGG and LGG tumor segmentation in clinical application.

Hypothesis Sketching for Online Kernel Selection in Continuous Kernel Space

Online kernel selection in continuous kernel space is more complex than that in discrete kernel set. But existing online kernel selection approaches for continuous kernel spaces have linear computational complexities at each round with respect to the current number of rounds and lack sublinear regret guarantees due to the continuously many candidate kernels. To address these issues, we propose a novel hypothesis sketching approach to online kernel selection in continuous kernel space, which has constant computational complexities at each round and enjoys a sublinear regret bound. The main idea of the proposed hypothesis sketching approach is to maintain the orthogonality of the basis functions and the prediction accuracy of the hypothesis sketches in a time-varying reproducing kernel Hilbert space. We first present an efficient dependency condition to maintain the basis functions of the hypothesis sketches under a computational budget. Then we update the weights and the optimal kernels by minimizing the instantaneous loss of the hypothesis sketches using the online gradient descent with a compensation strategy. We prove that the proposed hypothesis sketching approach enjoys a regret bound of order O(√T) for online kernel selection in continuous kernel space, which is optimal for convex loss functions, where T is the number of rounds, and reduces the computational complexities at each round from linear to constant with respect to the number of rounds. Experimental results demonstrate that the proposed hypothesis sketching approach significantly improves the efficiency of online kernel selection in continuous kernel space while retaining comparable predictive accuracies.