M. Fatih Akay

Predicting the performance measures of an optical distributed shared memory multiprocessor by using support vector regression

Recent advances in the development of optical technologies suggest the possible emergence of optical interconnects within distributed shared memory (DSM) multiprocessors. The performance of these DSM architectures must be evaluated under varying values of DSM parameters. In this paper, we develop a Support Vector Regression (SVR) model for predicting the performance measures (i.e. average network latency, average channel waiting time and average processor utilization) of a DSM multiprocessor architecture interconnected by the Simultaneous Optical Multiprocessor Exchange Bus (SOME-Bus), which is a high-bandwidth, fiber-optic interconnection network. The basic idea is to collect a small number of data points by using a statistical simulation and predict the performance measures of the system for a large set of input parameters based on these. OPNET Modeler is used to simulate the DSM-based SOME-Bus multiprocessor architecture and to create the training and testing datasets. The prediction error and correlation coefficient of the SVR model is compared to that of Multiple Linear Regression (MLR) and feedforward Artificial Neural Network (ANN) models. Results show that the SVR-RBF model has the lowest prediction error and is more robust. It is concluded that SVR model shortens the time quite a bit for obtaining the performance measures of a DSM multiprocessor and can be used as an effective tool for this purpose.

Performance evaluation of directory protocols on an optical broadcast-based distributed shared memory multiprocessor

Recent advances in the development of optical technologies suggest the possible emergence of broadcast-based optical interconnects within cache-coherent distributed shared memory (DSM) multiprocessor architectures. It is well known that the cache-coherence protocol is a critical issue in designing such architectures because it directly affects memory latencies. In this paper, we evaluate via simulation the performance of three directory-based cache-coherence protocols; strict request-response, intervention forwarding and reply forwarding on the Simultaneous Optical Multiprocessor Exchange Bus (SOME-Bus), which is a low-latency and high-bandwidth broadcast-based fiber-optic interconnection network supporting DSM. The simulated system contains 64 nodes, each of which has a processor, a cache controller, a directory controller and an output channel. Simulations have been conducted for each protocol to measure average processor utilization, average network latency and average number of packets transferred over the network for varying values of the important DSM parameters such as the ratio of the mean channel service time to mean thread run time (T/R), probability of a cache block being in modified state {P(M)}, the fraction of write misses {P(W)} and home node contention rate. The results reveal that for all cases, except for low values of P(M), intervention forwarding gives the worst performance (lowest processor utilization and highest latency). The performance of strict request-response and reply forwarding is comparable for several values of the DSM parameters and contention rate. For a contention rate of 0%, the increase of P(M) makes reply forwarding perform better than strict request-response. The performance of all protocols decreases with the increase of P(W) and contention rate. However, the performance of strict request-response is the least affected among other protocols due to the negative impact of the increase of P(W) and contention rate. Therefore, for the full contention case (i.e. contention rate of 100%); for low values of P(M), or for mid values of P(M) and high values of P(W), strict request-response performs better than reply forwarding. These results are significant in the sense that they provide an insight to multiprocessor architecture designers for comparing the performance of different directory-based cache-coherence protocols on a broadcast-based interconnection network for different values of the DSM parameters and varying rates of contention.

Predicting upper body power of cross-country skiers using machine learning methods combined with feature selection

Upper body power (UBP) is one of the most important determinants of cross-country ski race performance. In this study, General Regression Neural Networks (GRNN), Radial Basis Function Neural Network (RBF), Decision Tree Forest (DTF) combined with a feature selection algorithm have been used to developed prediction models for estimating 10-second UBP (UBP10) and 60-second UBP (UBP60) of cross-country skiers. By using the Relief-F attribute selection algorithm, the score of each attribute has been calculated. Seven different UBP10 and UBP60 prediction models have been developed by removing the attribute with the lowest score at a time. By using 10-fold cross-validation on the data set, the performance of the prediction models has been evaluated by calculating their multiple correlation coefficients (R) and standard error of estimate (SEE). The results show that gender and VO2max are the most effective variables for prediction of UBP10 and UBP60.

Development of new upper body power prediction models for cross-country skiers by using different machine learning methods

Upper Body Power (UBP) is one of the most important determinants that directly affects the performance of cross-country skiers during races. In this study, new models have been developed to predict the 10-second UBP (UBP10) and 60-second UBP (UBP60) of cross-country skiers by using different machine learning methods including Cascade Correlation Network (CCN), Radial Basis Function Neural Network (RBF) and Decision Tree Forest (DTF). The predictor variables used to develop prediction models are age, gender, body mass index (BMI), heart rate (HR), maximal oxygen uptake (VO2max) and exercise time. By using 10-fold cross-validation on the dataset, the performance of the prediction models has been evaluated by calculating their multiple correlation coefficient (R) and standard error of estimate (SEE). The results show that the CCN-based model including the predictor variables age, gender, BMI and VO2max yields the lowest SEE both for the prediction of UBP10 and UBP60.

Prediction of maximum oxygen uptake with different machine learning methods by using submaximal data

Maximum oxygen uptake (VO2max) is the highest amount of oxygen used by the body during intense exercise and is an important component to determine cardiorespiratory fitness. In this study, models have been developed for predicting VO2max with four different machine learning methods. These methods are Treeboost (TB), Decision Tree Forest (DTF), Gene Expression Programming (GEP) and Single Decision Tree (SDT). The predictor variables used to develop prediction models include gender, age, weight, height, treadmill speed, heart rate and stage. The performance of the prediction models have been evaluated by calculating Standard Error of Estimate (SEE) and Multiple Correlation Coefficient (R) and using 10-fold cross validation. Results show that compared to the SEEs of TB, the maximum percentage decrement rates in SEEs of DTF, GEP and SDT are 8.38%, 12.97% and 23.07%, respectively.

Performance comparison of different regression methods for maximal oxygen uptake estimation of cross-country skiers

Maximal oxygen uptake (VO2max) is the one of the most important determinants of cross-country ski race performance. The purpose of this study is to develop new VO2max prediction models for cross-country skiers by using General Regression Neural Network (GRNN), Cascade Correlation Network (CCN) and Single Decision Tree (SDT). In order to develop VO2max prediction models, a dataset including data of 139 subjects and the input variables age, gender, height, weight, body mass index (BMI), heart rate at lactate threshold (HRLT), maximum heart rate (HRmax) and time have been used. Applying 10-fold cross validation on the dataset, multiple correlation coefficients (Rs) and standard error of estimates (SEEs) of the models have been calculated. It is shown that GRNN-based models yield 12.13% and 25.50% lower SEEs on the average than the ones obtained by CCN-based and SDT- based models.

Determination of the variables affecting the maximal oxygen uptake of cross-country skiers by using machine learning and feature selection algorithms

Maximal oxygen uptake (VO2max) is one of the most important determinants that directly affects the performance of cross-country skiers during races. In this study, various models have been developed to predict the VO2max of cross-country skiers by combining different machine learning methods with the Relief-F feature selection algorithm. Machine learning methods used in this study include General Regression Neural Network (GRNN), Gene Expression Programming (GEP), Group Method of Data Handling Polynomial Network (GMDH) and Single Decision Tree (SDT). The predictor variables used to develop prediction models are age, gender, weight, height, heart rate (HR), heart rate at lactate threshold (HRLT) and exercise time. By using 10-fold cross-validation on the dataset, the performance of the prediction models has been evaluated by calculating their multiple correlation coefficient (R) and standard error of estimate (SEE). The results show that the GRNN-based model including all predictor variables yields the highest R (0.92) and the lowest SEE (2.98 ml kg-1 min-1).

Development of new non-exercise maximum oxygen uptake models by using different machine learning methods

Maximal oxygen consumption (VO2max) is the highest amount of oxygen used by the body during intense exercise. In this study, new non-exercise models have been developed by using different machine learning methods for predicting the VO2max values of healthy individuals aged between 18 and 65 years. The models include the non-exercise physiological variables (gender, age, weight and height) and questionnaire data. Cascade Correlation Network (CCN), Group Method of Data Handling (GMDH), Decision Tree Forest (DTF) and Single Decision Tree (SDT) methods have been used for developing the prediction models. The performance of the prediction models has been evaluated by calculating their multiple correlation coefficient (R) and standard error of estimate (SEE). The results show that CCN-based prediction models yield 24.54% on the average lower SEEs than the ones obtained by other methods.