Qatar Foundation Annual Research Conference Proceedings Volume 2016 Issue 1

Mobile learning can help in evolving students learning strength and comprehension skills. A connection is required to enable such devices communicate between each other. Brain-Computer Interface (BCI) can read brain signals and transform them into readable information. For instance, an instructor can use such device to track interest, stress level, and engagement of his students. We propose in this paper a mobile learning system that can transpose text-to-picture (TTP) to illustrate the content of Arabic stories and synchronize information with connected devices in a private wireless mesh network. The device of the instructor can connect the internet to download further illustrative information. It shares Wireless and Bluetooth connection with at least one student. Therefore, students can share the connection between each other to synchronize information on their devices. BCI devices are used to navigate, answer questions, and get to track students' performance. The aim of our educational system is to establish a private wireless mesh network that can operate in a dynamic environment.

Keywords

Mobile Learning, Arabic Text Processing, Brain Computer Interface, Engineering Education, Wireless Mesh Network.

I. Introduction

Nowadays mobile devices and collaborative work opened a new horizon for collaborative learning. As most people own handheld private portable smart phones, this has become the main means of connectivity and communication between people. Using smart phones for learning is beneficial and more attractive as learners can access educational resources at any time. Different eLearning systems available provide different options for collaborative classroom environment. However, they do not consider effective needs to adapt learning performance. They do not provide dynamic communication, automatic feedback, and other required classroom events.

II. Backgrounds

There are several collaborative learning applications have been proposed. BSCW [1] enables for online sharing of workspaces between distant people. Lotus Sametime Connect [2] also provides services for collaborative multiway chat, web conferencing, location awareness, and so. Saad et al. [3] proposed an intelligent collaborative system that can enable small range of mobile devices to communicate using WLAN, and uses the Bluetooth in case of power outage. The architecture of the proposed system is central where client can connect the server. Saleem [4] proposed a Bluetooth Assessment System (BAS) to use Bluetooth as an alternative to transfer questions and answers between the instructor and students. As many systems have been proposed in the domain of collaborative learning, several of them support mobile technology while others do not. BCI is not considered as part of the mobile educational system that can surround the environment with reading mental signals.

III. The proposed system

Our proposed system provides educational content and and synchronize it in a Wireless and Bluetooth wireless mesh network. It can be used in classrooms independent from the public network. The primary device broadcasts messages to enable users follow the explanation of the instructor on their mobile devices. The proposed system covers: 1) the establishment of wireless mesh network between mobile devices, 2) reading BCI data, 3) message communications, and 4) performance analysis between Wireless and Bluetooth technologies through device-to-device communication.

References

[1] K. Klö, “BSCW: cooperation support for distributed workgroups”, Parallel, Distributed and Network-based Processing. In 10th Euromicro Workshop, pp. 277–282, 2002.

[2] Lotus Sametime Connect. (2011, Feb. 17). Available: http://www.lotus.com/sametime.

[3] T. Saad, A. Waqas, K. Mudassir, A. Naeem, M. Aslam, S. Ayesha, A. Martinez-Enriquez, and P.R. Hedz, “Collaborative work in class rooms with handheld devices using bluetooth and WLAN,” IEEE 27th Canadian Conference on Electrical and Computer Engineering (CCECE), 2014, pp. 1–6.

[4] N.H., Saleem, “Applying Bluetooth as Novel Approach for Assessing Student Learning,” Asian Journal of Natural & Applied Sciences, vol 4, no. 2, 2015.

Motivation Chronic kidney disease (CKD) refers to the loss of kidney functions over time which is primarily to filter blood. Based on its severity it can be classified into various stages with the later ones requiring regular dialysis or kidney transplant. Chronic kidney disease mostly affects patients suffering from the complications of diabetes or high blood pressure and hinders their ability to carry out day-to-day activities. In Qatar, due to the rapidly changing lifestyle there has been an increase in the number of patients suffering from CKD. According to Hamad Medical Corporation [2], about 13% of Qatar's population suffers from CKD, whereas the global prevalence is estimated to be around 8–16% [3]. CKD can be detected at an early stage and can help at-risk patients from a complete kidney failure by simple tests that involve measuring blood pressure, serum creatinine and urine albumin [1]. Our goal is to use machine learning techniques and build a classification model that can predict if an individual has CKD based on various parameters that measure health related metrics such as age, blood pressure, specific gravity etc. By doing so, we shall be able to understand the different signals that identify if a patient at risk of CKD and help them by referring to preventive measures.

Dataset Our dataset was obtained from the UCI Machine Learning repository, which contains about 400 individuals of which 250 had CKD and 150 did not. The dataset was obtained from a hospital in southern India over a period of two months. In total there are 24 fields, of which 11 are numeric and 13 are nominal i.e. can take on only one of many categorical values. Some of the numerical fields include: blood pressure, random blood glucose level, serum creatinine level, sodium and potassium in mEq/L. Similarly, examples of nominal fields are answers to yes/no type questions such as whether the patient suffers from hypertension, diabetes mellitus, coronary artery disease. There was missing data values in a few rows which was addressed by imputing them with the mean value of the respective column feature. This ensures that the information in the entire dataset is leveraged to generate a model that best explains the data.

Approach We use two different machine learning tasks to approach this problem, namely: classification and clustering. In classification we built a model that can accurately classify if a patient has CKD based on their health parameters. And in order to understand if people can be grouped together based on the presence of CKD we have performed clustering on this dataset. Both these approaches provide good insights into the patterns present in the underlying data. Classification This problem can be modeled as a classification task in machine learning where the two classes are: CKD and not CKD which represents if a person is suffering from chronic kidney disease or not respectively. Each person is represented as a set of features provided in the dataset described earlier. We also have ground truth as to if a patient has CKD or not, which can be used to train a model that learns how to distinguish between the two classes. Our training set consists of 75% of the data and the remaining 25% is used for testing. The ratio of CKD to non-CKD persons in the test dataset was maintained to be approximately the similar to the entire dataset to avoid the problems of skewness. Various classification algorithms were employed such as logistic regression, Support Vector Machine (SVM) with various kernels, decision trees and Ada boost so as to compare their performance. While training the model, a stratified K-fold cross validation was adopted which ensures that each fold has the same proportion of labeled classes. Each classifier has a different methodology for learning. Some classifiers assign weights to each input feature along with a threshold that determines the output and updates them accordingly based on the training data. In the case of SVM, kernels map input features into a different dimension which might be linearly separable. Decision tree classifiers have the advantage that it can be easily visualized since it is analogous to a set of rules that need to be applied to an input feature vector. Each classifier has a different generalization capability and the efficiency depends on the underlying training and test data. Our aim is to discover the performance of each classifier on this type of medical information.

Clustering Clustering involves organizing a set of items into groups based on a pre-defined similarity measure. This is an unsupervised learning method that doesn't use the labeled information. There are various popular clustering algorithms and we use k-means and hierarchical clustering to analyze our data. K-means involves specifying the number of classes and the initial class means which are set to random points in the data. We vary the number of groups from 2 to 5 to figure out which maximizes the quality of clustering. Clustering with more than 2 groups also might allow to quantify the severity of Chronic Kidney Disease (CKD) for each patient instead of the binary notion of just having CKD or not. In each iteration of k-means, each person is assigned to a nearest group mean based on the distance metric and then the mean of each group is calculated based on the updated assignment. After a few iterations, once the means converge the k-means is stopped. Hierarchical clustering follows another approach whereby initially each datapoint is an individual cluster by itself and then at every step the closest two clusters are combined together to form a bigger cluster. The distance metric used in both the methods of clustering is Euclidean distance. Hierarchical clustering doesn't require any assumption about the number of clusters since the resulting output is a tree-like structure that contains the clusters that were merged at every time-step. The clusters for a certain number of groups can be obtained by slicing the tree at the desired level. We evaluate the quality of the clustering based on a well known criteria known as purity. Purity measures the number of data points that were classified correctly based on the ground truth which is available to us [5].

Principal Component Analysis Principal Component Analysis (PCA) is a popular tool for dimensionality reduction. It reduces the number of dimensions of a vector by maximizing the eigenvectors of the covariance matrix. We carry out PCA before using K-Means and hierarchical clustering so as to reduce it's complexity as well as make it easier to visualize the cluster differences using a 2D plot.

Results Classification In total, 6 different classification algorithms were used to compare their results. They are: logistic regression, decision tree, SVM with a linear kernel, SVM with a RBF kernel, Random Forest Classifier and Adaboost. The last two classifiers fall under the category of ensemble methods. The benefit of using ensemble methods is that it aggregates multiple learning algorithms to produce one that performs in a more robust manner. The two types of ensemble learning methods used are: Averaging methods and Boosting methods [6].

The averaging method typically outputs the average of several learning algorithms and one such type we used is random forest classifier. On the other hand, a boosting method “combines several weak models to produce a powerful ensemble” [6]. Ada boost is an example of boosting method that we have used.

We found that the SVM with linear kernel performed the best with 98% accuracy in the prediction of labels in the test data. The next best performance was by the two ensemble methods: Random Forest Classifier with 96% and Adaboost 95% accuracy. The next two classifiers were: Logistic regression with 91% and Decision tree with 90%. The classifier with the least accuracy was SVM with a RBF kernel which has about 60% accuracy. We believe that RBF gave lower performance because the input features are already high dimensional and don't need to be mapped into a higher dimensional space by RBF or other non-linear kernels. A Receiver Operating Characteristic (ROC) curve can also be plotted to compare the true positive rate and false positive rate. We also plan to compute other evaluation metrics such as precision, recall and F-score. The results are promising as majority of the classifiers have a classification accuracy of above 90%.

After classifying the test dataset, feature analysis was performed to compare the importance of each feature. The most important features across the classifiers were: albumin level and serum creatinine. Logistic regression classifier also included the ‘pedal edema’ feature along with the previous two features mentioned. Red blood cell feature was included as an important feature by Decision tree and Adaboost classifier.

Clustering After performing clustering on the entire dataset using K-Means we were able to plot it on a 2D graph since we used PCA to reduce it to two dimensions. The purity score of our clustering is 0.62. A higher purity score (max value is 1.0) represents a better quality of clustering. The hierarchical clustering plot provides the flexibility to view more than 2 clusters since there might be gradients in the severity of CKD among patients rather than the simple binary representation of having CKD or not. Multiple clusters can be obtained by intersecting the hierarchical tree at the desired level.

Conclusions We currently live in the big data era. There is an enormous amount of data being generated from various sources across all domains. Some of them include DNA sequence data, ubiquitous sensors, MRI/CAT scans, astronomical images etc. The challenge now is being able to extract useful information and create knowledge using innovative techniques to efficiently process the data. Due to this data deluge phenomenon, machine learning and data mining have gained strong interest among the research community. Statistical analysis on healthcare data has been gaining momentum since it has the potential to provide insights that are not obvious and can foster breakthroughs in this area.

This work aims to combine work in the field of computer science and health by applying techniques from statistical machine learning to health care data. Chronic kidney disease (CKD) affects a sizable percentage of the world's population. If detected early, its adverse effects can be avoided, hence saving precious lives and reducing cost. We have been able to build a model based on labeled data that accurately predicts if a patient suffers from chronic kidney disease based on their personal characteristics.

Our future work would be to include a larger dataset consisting of of thousands of patients and a richer set of features that shall improve the richness of the model by capturing a higher variation. We also aim to use topic models such as Latent Dirichlet Allocation to group various medical features into topics so as to understand the interaction between them. There needs to be a greater encouragement for such inter-disciplinary work in order to tackle grand challenges and in this case realize the vision of evidence based healthcare and personalized medicine.

References

[1] https://www.kidney.org/kidneydisease/aboutckd

[2] http://www.justhere.qa/2015/06/13-qatars-population-suffer-chronic-kidney-disease-patients-advised-take-precautions-fasting-ramadan/

[3] http://www.ncbi.nlm.nih.gov/pubmed/23727169

[4] https://archive.ics.uci.edu/ml/datasets/Chronic_Kidney_Disease

[5] http://nlp.stanford.edu/IR-book/html/htmledition/evaluation-of-clustering-1.html

[6] http://scikit-learn.org/stable/modules/ensemble.html

1. Introduction

Twitter is the micro blogging social media platform which has at most variety of content. The open access to Twitter data with the usage of Twitter APIs has made it a important area of research. Twitter has a useful feature called “Trends” which displays Hot topics or trending information differing for every location. This trending information is evolved by tweets being shared on Twitter in a particular location. But Twitter limits the trending information to current tweets, as the algorithm for finding trends is concentrated on generating trends in real time, rather than trends summarization of hot topics on daily basis. Thus a clear summarization of the contemporary trending information is missing and is much needed. Latest Twitter Trends - our application discussed in this paper, is built to get the aggregate of hot topics on Twitter for Arab Countries and the World. This is a real time application with summarization of Hot topics over time. It enables users to study the summarization of twitter trends by location with the help of a Word Cloud. The tool also enables the user to click on the particular trend, which will allow the user to navigate and search through Twitter Search - also in real time. This tool also overcomes a drawback of Twitter trending information, in addition to the Twitter trend algorithm. The trends differ for different languages in different locations and are often mixed. For eg, if #Eid-ul-Adha is trending in Arab countries, عيد الأضحى# is also trending. This application focuses on consolidating the trends in Arabic and English, which have the same meaning and display only one trending topic, instead of two same topics in different language. This application also gives an estimation of the different kind of Twitter users, analyzing the percentage of tweets made by Male and Female in that location.

2. Trends data gathering

Twitter APIs give developers access to the real time data comprising of tweets & trends. The Twitter REST API is used by the tool - Latest Twitter Trends, to connect and get trending data from Twitter. The API is used to authenticate and establish connection with Twitter and also returns Twitter trending data in JSON format. Python programming language is used to write scripts to gather data from Twitter. A Data Crawling Script is developed for connecting with Twitter API by authenticating the credentials generated by twitter on creating an application from the app.twitter.com. The Customer Key, Customer Secret, Access Token, Access Token Secret are the credentials used to perform authentication by Twitter. The data returned by Twitter is in JSON (JavaScript Object Notation) format and the Python Data Crawling Script is commanded to handle the JSON files and create a CSV database. This High Level Gathering of Data comprises of the following: Python data crawling script connects and authenticates with Twitter API and gets trending places data in JSON format from Twitter. The data in JSON format is stored in to our tool database as a CSV file. The Twitter data gathered is all the trending location/places with the WOIED (Where On Earth ID). The WOEID is used as a key to get Twitter trending topics location by location - in real time using the Twitter REST API. The trends for every location are also returned to the tool in JSON format, which is again changed converted to CSV for saving in the tool database. This CSV file for Trends is appended every time a new trending data is collected from twitter. Another CSV file is maintained in the Database which holds only the current information for all trending places - for later use. Natural language processing is done on trends by location CSV data, dictionary, to consolidate and consider Arabic and English Trending topics as one. The results are stored in CSV file and will be used for the hot topic identification.

3. Hot topic identification

After the High level Data Gathering, CSV files containing data are used as a Database for generating Word Cloud using D3.js. This trending data is processed by calculating the number of occurrences to give an estimate of which trending topic was trending for a long time. The frequency is taken as the count value for trending topic and a word cloud is generated. This algorithm for calculating frequency is a python script, written mainly for Word Cloud Data Crawling. This word cloud data crawling script takes the Trends by Location data as input and generates a huge database of trends by cities in JSON files. This word cloud crawling script gives output in the JSON files to be stored with key as the trend topic and value as the frequency of the trend occurrence.

4. Architecture

Figure 1: Latest Trends in Twitter Application Architecture The python scripts for data crawling and word cloud crawling are sued to connect with Twitter, gather data, process and store in a database. The D3.js and Google fusion table API are used for displaying the application results. Google Fusion Table API is used to create a Map containing current trends by location – geo-tagged on the map. Java program is used as a dedicated project to connect & authenticate with Google API and clear old fusion table data to import new updated rows in to the Google Fusion Table. Python script Tagcloud.py is used to generate cities. JSON with trending topics from the Trends.csv file. These files from the database for generating word cloud using D3.js, individually for every city/location. Fusion table is used to visualize the trending information from Twitter. A java program along with Google API is used to authenticate and connect. Also to delete previous information in fusion table and update/import new records of data.

5. Results

The data crawling script establishes connection with Twitter and returns a JSON format as in Fig. 2. This data is processed and saved as a CSV in to our application database for later use. Figure 2: Trends Data Output from Twitter in JSON format The word cloud crawling script generates key value pairs of processed trending data from the database. The key containing trending topic and the value containing the frequency of the trending topic's occurrence. The Fig. 3 displays the JSON dataset used for generating word cloud. Figure 3: JSON data of the processed trending data The word cloud is generated using the D3.js library and is used to display summarized trending data to the user. Figure 4 shows the word cloud result for London country. Figure 4: Word cloud for trending data.

Mobile-health (m-health) systems leverage wireless and mobile communication technologies to promote new ways to acquire, process, transport, and secure the raw and processed medical data. M-health systems provide the scalability needed to cope with the increasing number of elderly and chronic disease patients requiring constant monitoring. However, the design and operation of such pervasive health monitoring systems with Body Area Sensor Networks (BASNs) is challenging in twofold. First, limited energy, computational and storage resources of the sensor nodes. Second, need to guarantee application level Quality of Service (QoS). In this paper, we integrate wireless network components, and application-layer characteristics to provide sustainable, energy-efficient and high-quality services for m-health systems. In particularly, we propose an Energy-Cost-Distortion (E-C-D) solution, which exploits the benefits of medical data adaptation to optimize transmission energy consumption and cost of using network services. However, at large scale networks and due to heterogeneity of wireless m-health systems, centralized approach becomes less efficient. Therefore, we present a distributed cross-layer solution, which is suitable for heterogeneous wireless m-health systems and scalable with the network size. Our scheme leverages Lagrangian duality theory and enables us to find efficient trade-off among energy consumption, network cost, and vital signs distortion, for delay sensitive transmission of medical data over heterogeneous wireless environment. In this context, we propose a solution that enables energy-efficient high-quality patient health monitoring to facilitate remote chronic disease management. We propose a multi-objective optimization problem that targets different QoS metrics, namely, signal distortion, delay, and Bit Error Rate (BER), as well as monetary cost and transmission energy. In particularly, we aim to achieve the optimal trade-off among the above factors, which exhibit conflicting trends.

The main contributions of our work can be summarized as follows:

•   (1) We design a system for EEG health monitoring systems that achieves high performance by properly combining network functionalities and EEG application characteristics.
• (2) We formulate a cross-layer multi-objective optimization model that aims at adapting and minimizing, at each PDA, the encoding distortion and monetary cost at the application layer, as well as the transmission energy at the physical layer, while meeting the delay and BER constraints.
• (3) We use geometric program transformation to convert the aforementioned problem into a convex problem, for which an optimal, centralized solution can be obtained.
• (4) By leveraging Lagrangian duality theory, we then propose a scalable distributed solution. The dual decomposition approach enables us to decouple the problem into a set of sub-problems that can be solved locally, leading to a scalable distributed algorithm that converges to the optimal solution.
• (5) The proposed distributed algorithm for EEG based m-health systems is analyzed and compared to the centralized approach.

Our results show the efficiency of our distributed solution, its ability to converge to the optimal solution and to adapt to varying network conditions. In particular, simulation results show that the proposed scheme achieves the optimal trade-off between energy efficiency and QoS requirements of health monitoring systems. Moreover, it offers significantly savings in the objective function (i.e., E-C-D utility function), compared to solutions based on equal bandwidth allocation.

Every day, humans interact with text spanning from different sources such as news, literature, education, and even social media. While reading, humans process text word by word, accessing the meaning of a particular word from the lexicon, and when needed, changing its meaning to match the context of the text (Harley, 2014). The process of reading can induce a range of emotions, such as engagement, confusion, frustration, surprise or happiness. For example, when readers come across unfamiliar jargon, this may confuse them, as they try to understand the text.

In the past, scientists have addressed the emotion in text from a writer's perspective. For example the field of Sentiment Analysis, aims to detect the emotional charge of words, to infer the intentions of the writer. However, here we propose the reverse approach: detect emotions produced on readers while processing text.

Detecting which emotions are induced by reading a piece of text can give us insights about the nature of the text itself. A word-emotion detector can be used to assign specific emotions experienced by readers to specific words or passages of text. This area of research has never been explored before.

There are many potential applications to a word-emotion detector. For example, a word-emotion detector can be used to analyze how passages in books, news or social media are perceived by readers. This can guide stylistic choices to cater for a particular audience. In a learning environment, it can be used to detect the affective states and emotions of students, so as to infer their level of understanding which can be used to provide assistance to students over difficult passages. In a commercial environment, it can be used to detect reactions to wording in advertizements. In the remainder of this report, we detail the first steps we followed to build a word-emotion detector. Moreover, we present the details of our system developed during QCRI's 2015 Hot Summer Cool Research internship program, as well as the initial experiments. In particular, we describe our experimental setup in which viewers watch a foreign language video with modified subtitles containing deliberate emotion inducing changes. We analyze the results and provide discussion about the future work.

The Pokerface System

A Pokerface is an inscrutable face that reveals no hint of a person's thoughts or feelings. The goal of the ‘Pokerface’ project is to build a word-emotion detector that works even if no facial movements are present. To do so, the Pokerface system uses a unique symbiose of the latest consumer-level technologies such as: eye-tracking to detect words that are being read; electroencephalography (EEG) to detect brain activity of the reader; and facial-expression recognition (FER) to detect movement in a reader's face. We then classify the brain activity and facial movements detected into emotions using Neural Networks.

In this report, we present the details of our Pokerface system, as well as the initial experiments done during QCRI's 2015 Hot Summer Cool Research internship program. In particular, we describe the setup in which viewers watch a foreign language video with subtitles containing deliberate emotion inducing changes.

Methodology

To detect emotions experienced by readers as they read text, we used different technologies. FER and EEG are used to detect emotional reactions through changes in facial expressions and brainwaves, while eye-tracking is used to identify the stimulus (text) to the reaction detected. A video interface was created to run the experiments. Below we describe each of them independently, and how we used them in the project.

EEG

EEG is the recording of electrical activity along the scalp (Niedermeyer and Lopes da Silva, 2005). EEG measures voltage fluctuations resulting from ionic current flows within the neurons of the brain. EEG is one of the few non-intrusive techniques available that provides a window on physiological brain activity. EEG averages the response from many neurons as they communicate, measuring the electrical activity by surface electrodes. We can then use the brain activity of a user to detect their emotional status.

Data Gathering

In our experiments, we used the Emotiv | EEG EPOC neuroheadset (2013), which has 14 EEG channels plus two references, inertial sensors, and two gyroscopes. The raw data from the neuroheadset was parsed with the timestamps for each sample.

Data Cleaning and Artifact Removal

After retrieving the data from the EEG, we need to remove “artifacts” which are changes in the signals that do not originate from neurons (Vidal, 1977), such as ocular movements, muscular movements, as well as technical noise. To do so, we used the open source toolbox EEGlab (Delorme & Makeig, 2004) to remove artifacts and filtering (removing the 4–45 Hz line noise).

ERP Collection

We decided to consider remaining artifacts as random noise and move forward with extracting Event Related Potentials (ERPs), since all of other options we had found required some level of manual intervention. ERPs are the relevant sections of our EEG data with regards to stimuli and the subjects' reaction time. To account for random effects form the artifacts, we averaged the ERPs over different users and events. To do so, we used EEGlab's plugin ERPlab, to add events codes to our continuous EEG data based on stimulus time.

Events

Our events were defined as textual modifications in subtitles, designed to induce emotions of confusion, frustration or surprise. The time at which the subject looks at a word was marked to be the stimulus time (st) for that word, and the reaction time was marked to be st+800 ms, because we rarely see a reaction to a stimulus 800 ms after its appearance (Fischler and Bradley, 2006).

The ERPs were obtained as the average of different events corresponding to the same condition (control or experimental).

Eye-Tracking

An eye-tracker is an instrument to detect the movements of the eye. Based on the nature of the eye and human vision, the eye-tracker identifies where a user is looking by shining a light that will be reflected into the eye, such that the reflection will be captured by image sensors. The eye-tracker will then measure the angle between the cornea and pupil reflections to calculate a vector and identify the direction of the gaze.

In this project, we used the EyeTribe eye-tracker to identify the words a reader looked at while reading. It was set up in a Windows machine. Before an experiment, the user needs to calibrate the eye-tracker. Recalibration is necessary every time the user changes their sitting position. While the eye-tracker is running, Javascript and NodeJS were used to create a function that extracts and parses the data from the machine and prints it into a text file. This data includes the screen's x and y coordinates of the gaze; the timestamp; and an indicator of whether the gaze point is a fixation or not. The data is received at a rate of 60fps. The gaze points are used to determine which words are looked at at any specific time.

Video Interface

In our experiments, each user was presented with a video with subtitles. To create the experimental interface, we made different design choices based on previous empirical research. Therefore, we used Helvetica font, given its consistency across all platforms (Falconer, 2011), and used the font size 26 given that it improves the readability of subtitles on large desktops (Franz, 2014). We used Javascript to detect the location of each word that was displayed on the screen.

After gathering the data the experiment, we used an off-line process to detect the “collisions” between the eye-tracker gaze points and the words displayed to the user. To do so, we used both time information and coordinate information. The result was a series of words annotated with the specific time-spans in which they were looked at.

FER

Facial Expression Recognition (FER) is the process of detecting an individual's emotion by accessing their facial expressions in an image or video. In the past, FER has been used for various purposes, including psychological studies, tiredness detection, facial animation and robotics, etc.

Data Gathering

We used the Microsoft Kinect with the Kinect SDK 2.0 for capturing the individual's face. The data extracted from the Kinect provided us with color and infrared images, as well as depth data. However, for this project we only worked with the color data. The data from the Kinect was saved as a sequence of color images, recorded at a rate of 30 frames per second (fps). The code made use of the process of multithreading to ensure high fps, and low memory usage. Each image frame was assigned a timestamp in milliseconds, which was saved in a text file.

Feature Extraction

After having extracted the data from the Kinect, the images were processed so as to locate the facial landmarks. We tested the images with Face++ which is a free API for face detection, recognition and analysis. Using Face++, we were able to locate 83 facial landmarks on the images. The data obtained from the API was the name of the landmark along with it's x and y-coordinates.

The next step involved obtaining Action Units (AUs) by using the facial landmarks located through Face++. Action Units are the actions of individual muscles or groups of muscles, such as, raising the outer eyebrow, or stretching of lips etc (Cohn et al. 2001). To determine which AUs to use for FER, as well as how to calculate them, Tekalp and Ostermann's (2000) was taken as a reference.

Classification

The final step of the process was classifying the image frames into one of the eight different emotions - happiness, sadness, fear, anger, disgust, surprise, neutral and confused. We used the MATLAB Neural Network toolkit (MathWorks, Inc., 2015) and designed a simple feed-forward neural network with backpropagation.

Pilot results

EEG

In our pilot classification study, we used we experimented with the Alcoholism data used in Kuncheva and Rodriguez (2012) paper, from the UC Irvine (UCI) machine learning repository (2010) which contains ERP raw data for 10 alcoholic subjects and 10 sober subjects. We extracted the features using the interval feature extraction. Total of 96K Features were extracted from each subject's data. We Achieved around 98% accuracy with the training data.

FER

We experimented on three different individuals as well as several images from the Cohn-Kanade facial expressions database. It was found that the application had roughly 75–80% accuracy, and this accuracy could further be improved by adding more data into the training set. It was also observed that the classifier was more accurate when classifying certain emotions than some others. For example, the images depicting happiness were classified accurately more than images that depicted any other emotion. The classifier had difficulty distinguishing between fear, anger and sadness.

Conclusion

In this paper, we presented Pokerface, a word-emotion detector that can detect the emotion of users as they read text. We built a video interface that displays subtitled videos and used the EyeTribe eye-tracker to identify the word a user is looking at a certain time in the subtitles. We used the Emotiv Epoc headset to obtain EEG brainwaves from the user, and the Microsoft Kinect to obtain their facial expressions, and extracted features from both. We used Neural Networks to classify both the facial expressions and the EEG brainwaves into emotions. In the future.

Future directions of work include to improve the accuracy of the FER and EEG emotion classification components. Furthermore, the EEG results can be improved by exploring additional artifact detection and removal techniques. Furthermore, we want to integrate the whole pipeline in a seamless application that allows effortless experimentation.

Once the setup is streamlined, Pokerface can be used to explore many different applications to optimize users' experiences in education, news, advertizing, etc. For example, the word-emotion detector can be utilized in Computer-assisted learning to provide students with virtual affective support, such as detecting confusion and providing clarifications.

Device-to-device (D2D) communication, or the establishment of direct communication between nearby devices, is an exciting and innovative feature of the next-generation networks. D2D has specially attracted the attention of the research community because it is a key ingredient in creating Internet of Things (IoT): a network of physical objects that are embedded with sensors to collect data about the world and networking interfaces to exchange this data. IoT can serve a wide variety of application domains ranging from health and education to green computing and property management. Fortunately, processors, communication modules and most electronic components are diminishing in size and price, which allows us to integrate them into more objects and systems.

Today, companies such as Intel and Cisco are trying to make general purpose IoT devices available to everyday developers, thus expediting the rate at which the world is adopting IoT. Many of those devices are becoming more affordable and the internet is swamped with tutorials on how to build simple systems with them. However, despite the fundamental importance of effective D2D communication in IoT, we have a shortage of work that goes beyond standards being developed to accurately assess practical D2D communication performance of such devices.

We address this gap by studying different communication metrics on representative general purpose IoT devices and examining their impact in practical settings. We share our experiences assessing the performance of different communication technologies and protocol stacks on these devices in different environments. Additionally, we utilize this knowledge to improve “UpnAway”, an agile UAV Cyber-Physical-System Testbed developed at CMU Qatar, by enhancing communication between IoT devices attached to the drones for on-board processing and autonomous navigation control.

Measuring D2D Performance in IoT devices

We investigated the performance of Intel Edison (Fig. 1, a) and Raspberry-Pi (Fig. 1, b) devices because they are the two most representative general purpose IoT devices. The Intel Edison devices are equipped with a dual-core CPU, single-core microcontroller, integrated Wi-Fi, Bluetooth 4.0 support, 1 GB DDR and 4GB Flash memory and 40 multiplexed GPIO interfaces. Being so highly equipped, they are being rapidly adopted by a substantial segment of IoT developers worldwide. The Raspberry-Pi devices are the currently most widely used devices, so we used them for results comparison and validation. We made our measurements on D2D WiFi and Bluetooth links because, on both the Edison and Raspberry-Pi devices, they are the most user friendly communication interfaces. The Intel Edison devices come with built-in Wifi and Bluetooth interfaces that can be controlled through friendly GUI and the Raspberry-Pi devices can support these technology by simply plugging the corresponding USB dongles.

Our investigations involved accurately measuring RTT, Throughput, Signal Strength and reliability. The experiments involved:

Sending time-stamped packets to an echo-server then using the echo to calculate the round-trip-time (RTT), Exchanging files multiple times then calculating average time-delay over different distances (Fig. 2), (Throughput), Increasing distance between nodes and reading signal strength (Signal Strength), Transferring large files repeatedly to test reliability of different protocols (reliability).

In addition to the data collected, these investigations helped us make some unsettling observations about the Intel's Edison devices. First, we found a bug in configuring BlueZ, the linux implementation of the Bluetooth stack, over Yocto, the operating system running on the Edison devices: the RFCOMM, a reliable protocol by specification, link was unable to catch transmission errors. We suggest using TCP/IP over a BNEP bluetooth connection as a reliable alternative for RFCOMM when using the Edison Devices. Second, we observed that the RTT of WiFi D2D connection between Edison devices was significantly higher than that between the Raspberry-Pi. We suspect that this is attributed to the Edison's energy saving feature.

Investigating A Relevant Application: UpNAway

The second part of our research involved using D2D communication between Edison Intel devices to improve UpnAway. We chose this application because of its significance in improving cyber-physical systems which can contribute to various industrial areas such as aerospace, transportation, robotics and factory automation systems. UpnAway addresses the Cyber-Physical System community's unmet need for a scalable, low-cost and easy-to-deploy test-bed.

The UpnAway testbed, however, this testbed was centralized: The UAV Node Modules (Fig. 4) used to run on a central computer while streaming motion instructions to the drones. It is a given that a distributed system has higher scalability and stability than a centralized one. So we helped the Up and Away team leverage D2D communication to upgrade the testbed to a distributed version of itself. We did this by mounting an Edison device on each drone. Then we established D2D connections between the edisons, their respective drones and the central node. Now, instead of having all computation done in the central node, the central drone now only performs localization and informs each Node Module, now running on the Edison, of its drone's coordinates.

Future Work

In the future, we would like to study the energy consumption aspect of data transmission over D2D connections. Further, we are interested in measuring how these devices perform over different technologies in different environments: outdoors, indoors in an open corridor and inside cars, for example. Such measurements will contribute to the design of many developing IoT systems.

Data centers have an important role in supporting cloud computing services (such as email, social networking, web search, etc.) enterprise computing needs, and infrastructure-based services. Data center networking is a research topic that aims at improving the overall performances of the data centers. It is a topic of high interest and importance for both academia and industry. Several architectures such as FatTree, FiConn, DCel, BCube, and SprintNet have been proposed. However, these topologies try to improve the scalability without any concerns about energy that data centers use and the network infrastructure cost, which are critical parameters that impact the performances of data centers.

In fact, companies suffer from the huge energy their data centers use and the network infrastructure cost witch is seen by operators as a key driver for maximizing data centers profits and according to industry estimates, the united states data center market achieved almost US\$39 billion in 2009, growing from US\$16.2 billion in 2005 Moreover, the studies show that the installed base of servers has been increasing 12 percent a year, from 14 million in 2000 to 35 million in 2008. Yet that growth is not keeping up with the demands placed on data centers for computing power and the amount of data they can handle. Almost 30 percent of respondents to a 2008 survey of data center managers said their centers will have reached their capacity limits in three years or sooner.

The infrastructure cost and power consumption are the first order design concern for data center operators. In fact, they represent an important fraction of the initial capital investment while not contributing directly to the future revenues. Thus, the design goals of data center architectures seen by operators are high scalability, low latency, low Average path length and especially low energy consumption and low infrastructure cost (the number of interface cards, switches, and links).

Motivated by these challenges, we propose a new data center architecture, called VacoNet that combines the advantages of previous architectures while avoiding their limitations. VacoNet is a reliable, high-performance, and scalable data center topology that is able to improve the network performances in terms average path length, network capacity and network latency. In fact, VacoNet can connect more than 12 times the number of nodes in FlatNet without increasing the APL. Also, it achieves a good network capacity even with a bottleneck effect (bigger than 0.3 even for 1000 servers). Furthermore, VacoNet reduced the infrastructure cost by about 50%, and the power consumption will be decreased with more than 50000 watt compared to all the previous architectures.

In addition, and thanks to the proposed fault tolerant algorithm, the new architecture shows a great performance even when the failure rate equals to 0.3, which means when about one third of the links failed, the connection failure rate is only 15%. By using VacoNet, operators can win till 2 million US dollars compared to Flatnet, Dcell, Bcube and Fattree.

Both theoretical analysis and simulation experiments have conducted and validated to evaluate the overall performance of the proposed architecture.

Qatar ratified the Convention on the Rights of Persons with Disabilities (CRPD) on May 13 2008, and has signed its Optional Protocol, pending ratification.

All persons with disabilities, and to promote respect for their inherent dignity aim the CRPD at promoting the full and equal enjoyment of all human rights and fundamental freedoms. The Optional Protocol (OP) establishes a complaints mechanism. The CRPD creates no new rights for the individuals, but makes it possible the full exercise of human rights to all. Complying with the CRPD requires the states to introduce numerous adjustments to their internal legislation.

The study of high significance for Qatar. At any rate, Qatar has being doing significant efforts, and major progress in catering to the needs of people with disabilities. Already in 1995, Qatar issued the Law No. 38/1995, on aspects of the Social Security system, providing governmental assistance to social groups, including organizations of persons with disabilities. In the 1998, Qatar created the Supreme Council for Family Affairs (SCFA, Decree No. 53/1998), a high-level national body that among other things has the mandate to deal with the implementation of those international conventions related to the rights of children, women, and persons with disabilities which has been ratified by Qatar. Following the SCFA's recommendations, Qatar passed in 2004, Qatar passed the Law No. 2/2004, for the protection of people with special needs, which ensures the rights of persons with disabilities in all fields.

According to that law, the people with special needs enjoy special protection in the State of Qatar, by means of:

 - Special education, health treatment, disease prevention and vocational training;

 - Receiving all the tools and means to facilitate their learning and mobility process;

 - Receiving special qualifications and training certificates upon completion of certain training programs and appoint them in areas that would accommodate their obtained skills and training;

 - Dedicating around 2% of the jobs in the private sector to people with special needs without any discrimination based on disability.

Nevertheless, the implementation of these efforts is till a work in progress. A United Nations Special Rapporteur on Disability, reported after a brief mission in Qatar that there is “a clear commitment from Qatari society to the needs of persons with disabilities”, which are tangible at Shafallah Centre for Children with Special Needs and at Al Noor Institute for the Blind. The rapporteur stressed that “”it appears that there is a clear commitment from the State and the private sector toward the issues confronting persons with disabilities in Qatar. Anecdotal evidence suggests that the private sector is a big contributor to institutions [for people with disabilities]”. Nevertheless, the rapporteur warned that “it also became clear that much of the caring and development remain almost exclusively disability-specific as opposed to the mainstreaming of the development needs of persons with disabilities. There appears to be a distinct lack of mainstreaming of disability in Qatar”[1].

In 2010, the International Disability Alliance, the global network aimed at promoting the effective and full implementation of the CRPD, recommended that “Qatar adopt a proactive and comprehensive strategy to eliminate de jure and de facto discrimination on any grounds and against all children, paying particular attention to girls, children with disabilities”[2]

The significance of this project comprehends several facets. This project is aimed at helping fulfill Qatar's commitment as requested by the CRPD. Beyond that, and above all, this project would have the direct potential to yield tangible benefits for people with disabilities in Qatar, by helping remove barriers that may prevent their full integration into mainstream society might made harder their personal and professional development. Moreover, this project should help the people with disabilities to become visible in Qatar.

The objectives of the proposed study are:

 To analyze the impact that the ratification of the CRPD has on the Qatari legal system;

 To elaborate a general framework aimed at defining possible ways in which the Qatari legal system could better develop the mandates of the CRPD;

 To elaborate recommendations about possible modifications to the internal Qatari legislation in order to specifically incorporate the mandates of the CRPD into the law of the country.

According to the Census 2010, carried out by the Qatari Statistics Authority, the total number of people with disabilities in Qatar is 7,743, which represents 0.45% of the total population of 1,699,435 inhabitants. Among non-Qataris, the percentage of people with disabilities is of a 0.28%, while among Qatari nationals the figure is six times higher, 1,71%, with 2972 persons with disabilities on a population of 174,279. All these numbers appear as significantly low by international standards, where people with disabilities compose around 10% of the population. Some phenomena could help explain these numbers, though [3].

The significantly low number of non-Qataris with disabilities can be explained by the fact that that group of population are mostly young, healthy workers who come to the country with a work contract, and after having passed medical tests of aptitude for the position. The disability among Qataris would, thus, reflect better the natural rate of disability in the country. However, this rate is still too low for international standards. The causes to this low number could be multiple. First, it may show that in Qatar there is not yet a full awareness about what constitutes a disability, and thus people do not declare themselves–or declare family members–as disabled. Second, it is possible that the tightly knitted family structure provides for the disabled, and thus little or no support is requested for outside the family, causing a considerable under-registration of cases of disabilities, since the disabled do not reach out for specific social or health services. Third, it is also possible that despite all efforts, Qatar still has a relatively poor network of services aimed at satisfying the needs of the disabled, who that way go under-detected in the official statistics. At any rate, the numbers show that there is significant room in Qatar for the development of strategies aimed at raising awareness and implementing programs and services for the disabled.

In that sense, this project might produce very valuable information, and propose high-impact measures to further the protection that Qatar provides to people with disabilities. This project should have, thus, a critical significance to people with disabilities in Qatar, and also to their families.

The Qatari population at large would benefit from this project, since it is ultimately aimed at helping integrate a group of people–the disabled–whose contribution to the country's human capital can be of extreme value, in a context of a growingly complex, diverse global society.

Finally, this project can become a valuable tool to help reaffirm Qatar's leadership in the region in matters of human rights and human development.

[1] Report of the Special Rapporteur's Mission to Qatar - Preliminary Observations (9 – 13 March 2010). http://www.un.org/disabilities/documents/specialrapporteur/qatar_2010.doc

[2] IDA CRPD FORUM Suggestions for disability-relevant recommendations 7th UPR Working group session (8 to 19 February 2010). http://www.internationaldisabilityalliance.org/sites/disalliance.e-presentaciones.net/files/public/files/UPR-7th-session-Recommendations-from-IDA.doc

[3] Qatar Statistics Authority. Census 2010. http://www.qix.gov.qa

Contemporary Qatari women's social progress can in part be measured through an analysis of current marriage practices. The Islamic marriage contract is a legal and religious document wherein Muslim brides indicate their expectations for post-marital life and an essential step in the marriage negotiation process. The conditions they stipulate in their marriage contracts are symbolic of the degree to which they exercise agency in their personal and professional lives as wives. As part of a larger study on marriage practices in Qatar, we collected and analyzed marriage contracts from a broad range of Qatari families. We treated these documents as archival evidence, reflecting changing bridal expectations from 1975–2013. A content analysis of contracts in our sample demonstrated an increase in the age at marriage for both Qatari men and women. The contracts also show the major areas in which brides negotiated the terms of their married lives including educational, professional, and household expectations. We read these stipulating conditions as moves to guarantee autonomy as wives.

Water Resources and Use in Qatar Prior to the Discovery of Oil Qatar is known as one of the most arid countries in the world with all of the land characterised as dessert which in geographical terms is defined as territory with no surface water. Despite this unpromising situation people have lived on the Qatar peninsula for thousands of years carefully conserving, harvesting and using the scarce water resources. This paper will argue that a better understanding of the conservation and utilisation of fresh water in the past may have implications for the future development of agriculture and water management in Qatar. All the water resources in the past were fed either directly by scarce and sporadic rainfall events or through sub-surface water (fossil water) which formed a fresh water lens floating above a predominantly saline aquifer. The presence of shallow wad is located throughout the country but particularly near the coast are indicative of the occasionally heavy rainfall. This paper will investigate the methods used to provide water to the traditional settlements and more ancient archaeological sites throughout the country. The paper will begin by reviewing the hydrological structure of the Qatar peninsula and its condition prior to the modern over exploitation of water resources from the 1950's onwards. The paper will pay particular attention to the location of settlements around the northern coast where particularly favourable conditions exist as a result of the north Qatar arch. The settlements of the northern coast employed a variety of water catchment methods including modifying shallow natural depressions (rawdhas) to provide either grazing land or in some cases agricultural zones. In some cases such as at Jifara extensive sunken field systems enclosed within mud brick walls were created fed by a series of shallow wells taking advantage of both rainfall catchment and fossil rainwater. In some cases such as at Ruwayda on the north Qatar coast an existing rawdah appears to have been modified to create a garden with trees enclosed by a fortified masonry wall. A variety of well forms were created including shaft wells where water was accessed by buckets or other receptacles lowered by rope (two example of traditional leather buckets with goat hair rope have been found in archaeological excavations in Qatar) and wide shallow stepped wells which allowed direct access to the water either for humans or animals. In the centre of the country access to water has always been more difficult because the freshwater acquifer has been closest to the surface near the coast. As a consequence the majority of human settlement in Qatar has always been close to the coast. Where inland settlement do exist they are usually located either within or next to some geographical feature such as a wadi or a large depression which will act as a catchment area for rainfall. Such settlements are nearly always supplemented with wells which tend to be larger and much deeper than those on the coast. Examples of inland wells include those associated with gardens at Umm Salal Muhammad. The transformation of inland wells with motorised pumps from the 1950's onwards was one of the causes for the depletion and subsequent salinization of the freshwater aquifer in Qatar. In addition to wells and modified rawdahs a number of other forms of water catchment exist. One of the most surprising water sources is the perennial spring (naba‘a) which following periods of heavy or sustained rainfall may result in water literally spring out from the ground. Another unusual source are fresh water springs located off the coast which were traditionally exploited by fishermen and pearl divers. A rare form of water catchment exists in the Jabal Jassasiya rock formations on the north-east coast of Qatar where the natural contours of the rock are modified to form a cistern blocked on one side by a masonry dam. Masonry dams- possibly of medieval date- have also been documented near Umm Salal Muhammad.