Swarm robotic experiments need a large number of robots with expensive hardware setup and a well-controlled environment. As an alternative, there are many swarm robotics simulators run on virtual environments but would not be the same as real-world experiments. The purpose of this research is to introduce a Mixed Reality simulator, by merging the realities and modeling the swarm behaviors in such a way that both real and virtual robots co-exist and interact with each other. In this approach, swarm behavioral experiments can be conducted with a few physical robots and an unconstrained amount of virtual robot instances. Virtual robot instances were implemented by imitating the characteristics of physical robots. The simulator application will provide the additional necessities and support interaction between realities. Moreover, existing physical robots can have virtual sensors and modules which are expensive and impossible in reality by use of Mixed Reality. This study has proven that the traditional limitations of swarm robotics can be overcome with the aid of Mixed Reality technologies in a systematic and generalized way.

Genome sequencing and genome assembly are the computational process of converting the sequence composition of the gene within the cell of an organism in a human readable form. Mitochondria is an important genome in the cell and there is a need to study this genome for various reasons. The process of determining the order of bases A,G,T,C in the genome is known as genome sequencing.While sequencing the genome the original genome is separated into huge number of small parts known as reads and the end results of sequencing is a huge pool of reads(strings of A,G,T,C). These reads must be assembled back in a computer so that a biologist can identify and annotate the functionality of it. There are several techniques were developed to sequence the genome, the modern approach is next generation sequencing. Ilumina sequencing is one of the broadly used next generation sequencing method and this method produce large number of high precision sequencing short reads whereas other older methods produce longer reads.So the computational complexity of assembling back this large amount of read is high but cost efficient. Here we mainly discuss on low coverage sequencing data assembly.The sequencing data consist of multiple copies of same genome in low coverage data the number of copies are relatively lower than high coverage data. In this project we examined the tools used for mitochondrial genome assembly by assembling different datasets and measured the parameters that make impact in the assembly process. From the results we obtained from the experiment we made decisions of doing mitochondrial genome assembly.

Alzheimer’s disease is recognized as one of the common diseases found among elders, which still has no successful cure. Different technologies such as micro array technology, Sanger sequencing, and Next Generation Sequencing have been used by various researchers for gathering samples. Out of these, Next Generation Sequencing has become more common nowadays, as it is a powerful platform which enables to sequence thousands or millions of DNA molecules simultaneously. A set of samples collected using Next Generation Sequencing technology is used in this study. In this study, our goal is to determine the best set of microRNA biomarkers which are highly deferentially expressed in Alzheimer’s disease. Initially, the data set is preprocessed with the aid of the Galaxy tool and python programming language. Significance value, fold change and area under curve analysis are the statistical methods which are used in this study. Random Forest algorithm and Principal Component Analysis are used for selecting the best set of biomarkers out of the data set obtained at the end of statistical analysis. Using the statistical methods, followed by machine learning techniques, we establish 25 microRNAs as biomarkers for Alzheimer’s disease. Furthermore, we provide an analysis of the selected 25 microRNAs with area under the receiver operating curve and classification algorithms.

In Olympic-style weightlifting athlete attempts to lift the weight plates on a barbell and scores are determined by a combination of the successful highest weight achieved in snatch and the clean-and-jerk actions. However, the current method does not objectively measure the player techniques. In this paper, we introduce a novel method to objectively measure player performance on weightlifting using human action recognition in videos. We introduce a method to assess player techniques in weightlifting by using skeleton-based human action recognition. In order to achieve our goal, we further introduce a new video dataset for action recognition in weightlifting which is annotated to frame level and introduce an automated scoring system through action recognition. We conclude our paper with qualitative and quantitative experimental results using non-Olympic players and 2016 Gold, Silver, and Bronze medalists in different weight categories (both men and women).

Artificial Intelligence has impacted in a variety of industries, leading the world towards revolutionary applications and services that are primarily driven by high-performance computation and storage facilities in the cloud. This is mainly due to the advantage of having higher computational power, larger storage capacity and scalability. But with the increase of millions of IoT devices, a huge amount of data is being generated by end devices. To process such data, the distributed end devices have to communicate with the cloud servers making it difficult to generate real-time decisions though it consumes a lot of resources including bandwidth, processing power, and storage facilities at the cloud. On the other hand, Edge computing architectures enable a distributed way to process data near the sources of data which leads to facilitate real-time processing. But with the limited resources in the end devices, it is quite challenging to perform complex AI algorithms. Hence to facilitate such services and to enable real time processing at the edge, a novel approach is proposed base on distributed computation, vectorization, computation offloading, parallelization, and federated learning techniques.

Internal Transcribed Spacer (ITS) region is vastly preferred among all genomic regions for phylogenetic studies associated with various plant species. Isolation of ITS regions from skim sequencing data results in more accurate inter-species as well as intra-species diversity analysis. Most of the previous studies utilized the available tools and pipelines to isolate fungal ITS regions from Illumina sequences. Botanists find it much difficult to gather plant ITS regions from skim sequencing data due to the lack of an efficient existing workflow. Our study focuses to come up with a workflow that comprises a user-friendly pipeline for the botanists to isolate plant ITS regions from Illumina skim sequencing data as accurately and as efficiently as possible for phylogenetic studies.

Well-known authentication mechanisms such as Public-key Infrastructure (PKI) and Identity-based Public-key Certificates (ID-PKC) are not suitable to integrate with the peer-to-peer (P2P) network environment. The reason is the difficulty in maintaining a centralized authority to manage the certificates. The authentication becomes even harder in an anonymous environment. We present three authentication protocols such that the users can authenticate themselves in an anonymous P2P network, without revealing their identities. Firstly, we propose a way to use existing ring signature schemes to obtain anonymous authentication. Secondly, we propose an anonymous authentication scheme utilizing secret sharing schemes. Finally, we propose a zero-knowledge-proof-based anonymous authentication protocol. We provide security justifications of the three protocols in terms of anonymity, completeness, soundness, resilience to impersonation attacks, and resilience to replay attacks. We hope our article will open up a useful topic for further research in the field of communication.

Non-invasive electroencephalogram (EEG) based Brain-Computer Interface (BCI) systems have been an interesting research area for many fields. However, most of the research done on this subject is synchronous, therefore the state of mind of the user is not similar to its natural behaviour. Considering to provide possible experience in practical applications, self-paced BCI systems started gaining popularity in recent years. However, there are certain challenges yet to be addressed when following this method. Out of the research done on self-paced BCI systems, most of them are focused on motor-imagery control whereas research on non-motor imagery mental tasks is limited. In this research, we analyse the possibility of using the techniques used in the motor-imagery method for non-motor imagery mental tasks to be fed into virtual object controlling applications. Research was done with 5 different classification models with the use of features from Fast Fourier Transform (FFT) and Wavelet Transform (WT). K-nearest neighbor model with features obtained with FFT sustained its performance continuously with a 0.56 cross validation value.

Chloroplast genes and genomes play an important role in plant phylogeny and species identification. Skim sequencing is getting low coverage genome sequencing data that has nuclear, chloroplast and mitochondria genome sequences. Since the fast development of high throughput sequencing technologies, it’s low cost to urge the low coverage data of the whole genome (usually concerning 20-30GB data), that is enough to assemble a whole chloroplast genome. To date, there are several assembly processes/ pipelines designed to assemble a whole chloroplast genome. However, what proportion knowledge is required or really utilized in such analysis is a problem. Having such information can facilitate biologists to style their experiments properly and cost-effectively. Biologists expect a straightforward, quick and easy procedure to assemble and annotate a circular chloroplast genome from Illumina NGS data. In this project, we’ll analysis the present procedures for chloroplast genome assembly and annotation, and work on developing the strategies to spot and choose the best set(s) of data and the procedure(s) to assemble a given chloroplast genome as accurately and efficiently, by statistical, computational and heuristic strategies.

Human face recognition and feature extraction allow people to recognize a huge number of faces in a small amount of time, rather than recognizing each image and it’s features individually through a normal human’s eyes. Using these technologies researches are being carried out to find the characters that look like humans. The face detection and feature extraction methods for human images are hardly applied to the cartoon images because the features of cartoon characters differ from human features. This research was conducted to find the techniques to face detection, feature extraction of cartoon characters and recognize look-alike cartoon characters for a given human image. We have created a Disney cartoon repository including 800 images from 77 characters, including 35 labeled landmarks using imglab tool. For cartoon face detection and feature extraction, landmark based models are trained using our dataset. Total 92 features (50 areas and 42 distances between landmarks) are stored as csv files along with the cartoon images. To compare features of a real image with all the cartoon image features euclidean distance was considered. To increase the performance, we used the landmark based model with a hair extraction model and also included a gender prediction model. Alternatively, we implemented a classification model to find the best matching cartoon character. It shows 84% accuracy on training data and 80% accuracy on validation after 100 epochs. Evaluating the results of each model is done by rank based comparison. As this is the beginning of human-cartoon mapping, we were able to find the doppelganger cartoon for a human with considerable accuracy and our own created cartoon dataset will be more useful to future researches.

Nanopore sequencing is a third-generation sequencing technology that can read long DNA or RNA fragments in real-time. Nanopore sequencers measure the change in the electrical current when nucleotide bases translocate through a protein nanopore. These signal level changes are utilized in various nanopore data analysis workflows (such as identifying DNA methylation, polishing and variant detection) to obtain useful results from nanopore sequencing data. Adaptive Banded Event Alignment (ABEA) is a dynamic programming algorithm that is used as a key component in many nanopore data analysis workflows. Prior investigations have shown that ABEA consumes 70% of total CPU time in Nanopolish, a popular nanopore data analysis software package. Thus, optimizing the ABEA algorithm is vital for efficient nanopore data analysis. A previous study has proposed an accelerated version of ABEA on GPUs using CUDA that improves the execution time, at the cost of higher energy consumption. With the advancements of High-Level Synthesis (HLS) tools, Field Programmable Gate Arrays (FPGAs) are becoming more and more popular as accelerators that are energy efficient. In this work, we explore the use of the OpenCL for accelerating ABEA on FPGA with energy considerations. We propose a modified version of ABEA for FPGAs using OpenCL and apply various optimization techniques, leading to a few different implementations. We compare the performance of our implementations with other implementations on different hardware platforms in terms of execution time and energy consumption. We show that our best implementation archives an energy consumption of only 43% of the previous implementation of ABEA on GPU, however, with around 4x increase in execution time.

Finding a satisfying e-resource that consists of legal documents in Sri Lanka is difficult. On behalf of making good platform to facilitate lawyers and other people who works with legal documents to search and refer legal documents, it is a vital task to extract keywords in the documents. With that aim we did this research to identify keywords in legal articles using machine learning techniques. This paper presents a comprehensive analysis of keyword extraction in legal domain by providing existing analysis , related works are also concerned. Because nowadays there are a lot of legal documents offered in electronic format. Therefore, it is easy for legal scholars and professionals to search and refer details of legal documents. The objective of this analysis is to explore an associate economical way to implement an algorithm to identify keywords in legal articles in Sri Lanka. If manual keyword extraction happens and if there are ten thousand documents to upload the system, and it takes ten minutes to extract core details manually for a particular document. Then for a person who works eight hours per day ,he needs two hundred nine days to complete this task. Thus the system to analyze this legal knowledge will serve effectively for lawyers and law students, which might address a lawyer’s role and may even become powerful to unleash such a task in the future. Designers of such systems face a key challenge that the bulk of those documents are in natural language streams which are lacking formal structure or different specific linguistics information. During this analysis, an associate unsupervised learning approach for automatically distinguishing necessary details in each legal document is extracted. The machine learning and deep learning algorithms based mostly analysis systems apply these strategies in the main for the document classification. Legal document classification, translation, account, data gathering are part of the goals obtained from this research. During this study, we tried to implement two different approaches and compare the success of keyword extraction. This paper is structured as follows. Introduction addressed the problem and some related works on the keyword extraction, then in next sections the methodology of our research and the results evaluation and finally the conclusion of our study is presented. Index Terms—keyword extraction, Machine Learning, Text Rank,TF-IDF, Legal Articles"

Machine Learning (ML) has contributed to many advances in science and technology. Recently a trend of applications in high-stake decision-making has been initiated. The advancement of ML made the decision-making process unclear with complex black-box models, especially the state-of-the-art models which have maximized the performance are more complex, inexplicable, and hard to explain. On the contrary, high-stakes settings as healthcare, finance, and criminal justice, have strict ethical concerns that made a mandatory requirement to explain each decision or the model as a whole. Besides, the acts and regulations like General Data Protection Regulation (GDPR) make it obligatory to explain the decisions made by computer systems and created a social right for explanations. One of the most pressing problems the field is explainability and interpretability of artificial intelligent systems. Moreover, it is necessary to ensure the fairness and transparency of a decision to obtain the stakeholders' trust. The theoretical knowledge of explainable machine learning is not well-tested on real-world problems with direct social impact. In this paper, we have identified a quandary that reflects the characteristics of a high-stakes machine learning problem in the public sector. A solution of an early warning system to predict the projects that could be unfunded in an educational crowdfunding platform in a resource-constrained environment has been presented.

Identifying hand gestures using surface electromyography (sEMG) signals is vital in the development of next-generation human-machine interfaces (HMI). sEMG based HMIs provide users with a more natural and convenient way to communicate with computing systems. sEMG signals recorded from muscle tissues give information about the intended muscle movements triggered by the brain waves. Identifying these movements allows developing interfaces that can control computing devices. In this research, an attempt was made to improve a hand gesture recognition model that could be used as a human-machine interface using an online open dataset of sEMG signals. First sEMG signals were preprocessed using a bandpass filter and notch filter to remove noises in the signal. Then various time, frequency, and time-frequency domain features extracted and they were fed into machine learning algorithms such as random forest, support vector machines (SVM), K-nearest neighbors (K-NN), and recurrent neural networks. All the results were validated using 10-fold cross-validation. Maximum testing accuracy of 90.03% was obtained using an SVM classifier with root mean square, mean frequency, and median frequency of the signal as features for 24 channel data. Later an attempt was also made to use this result to control a simple game developed in Unity using sEMG signals collected from an 8-channel signal acquisition device. "

To deliver real-time IoT services and applications, distributed computation and AI processing at the edge have been identified as an efficient solution compared to cloud-based paradigms. These solutions are expected to support the delay-sensitive IoT applications, autonomic decision making, and smart service creation at the edge in comparison to traditional IoT solutions. However, existing solutions have limitations concerning distributed and simultaneous resource management for AI computation and data processing at the edge; concurrent and real-time application execution; and platform-independent deployment. Hence, first, we propose a novel three-layer architecture that facilitates the above service requirements. Then we have developed a novel platform and relevant modules with integrated AI processing and edge computer paradigms considering issues related to scalability, heterogeneity, security, and interoperability of IoT services. Further, each component is designed to handle the control signals, data flows, microservice orchestration, and resource composition to match with the IoT application requirements. Finally, the effectiveness of the proposed platform is tested and verified.

Online examinations require human invigilation either by live monitoring or inspection of recorded video to ensure academic integrity. This process is not feasible always since exams can be taken at any time and it involves high cost. This paper proposes an Online Proctoring System to automate invigilation processes by making use of inputs from a web browser, without using any external hardware or standalone application. As a result, several misconducts during the examination can be identified and labeled as suspicious.

Software Defined Network (SDN) is a vastly evolving network architecture approach that physically separates the network control plane from the data plane. The interconnection devices take forwarding decisions solely based on a set of multi-criteria policy rules defined by external applications called controllers. Vehicular communication becomes a trending topic in the transportation technology. Most of the researchers pay attention to the development of vehicular communication using wireless Software Defined Network in order to overcome from issues such as lack of intelligence and scalability of the network. The recent developments in SDN have paved the way to control and manage wireless ad-hoc networks. Wireless ad-hoc networks are distributed networks that work without fixed infrastructure and in which each network node is willing to forward network packets for other network nodes which in return provides a reliable platform in vehicle-to-vehicle communication. The objective of this research is to review some benefits of wireless SDN and how SDN can be used to implement network on vehicle to vehicle communication.

Near-Infrared spectroscopy is used for better vein visualization to make the venipuncture process more efficient. There exist a few models that use the said mechanism. However, they are costly, have accuracy and availability issues, and are limited only to certain types of skin tones. Our objectives were to develop a low-cost mechanism of obtaining near-infrared spectroscopy by using the image-guided technique, low-cost hardware, optimized algorithms, and evaluate its efficiency and usefulness by a clinical trial. We have tested the prototype using different combinations of light sources with different intensities and have analyzed the results. To quantitatively analyze, we have compared the number of visible veins under high intensity and low intensity. The number of visible veins is either same or up to 5 veins higher when 18W is used compared to when an intensity of 60W is used. We have also observed that the darker skin tones that have zero visible veins at normal sight result up to 2-3 veins when the prototype is used. The number of veins increased from 1 to 5, when the device was used, on fairer skin as well. We plan to conduct a clinical trial and test the device on human subjects and get feedback from the end-users, and improve the prototype accordingly.

Most of ECG analysis based emotion recognition studies use different machine learning techniques. Main problem with these methods is lack of accuracy in classifying various emotions. The proposed method uses a large public dataset to increase accuracy and implements a Convolutional Neural Network to identify emotions. ECG data signals are preprocessed to increase the number of instances and important features are extracted using feature extraction methods and then features are fed to the CNN. Three CNN models are trained to predict the valence, arousal and the dominance values of the ECG signal, which are used to finalize the emotion by mapping those values to the valence-arousal-dominance 3D plane. The classification CNN models implemented in this proposed method result in a maximum accuracy of 80.87%.

With the availability of high performance processors and GPUs, the place for Machine learning , Deep learning algorithms are growing exponentially. It has become more and more possible to explore the depths of fields like Computer vision with these trends. Detecting humans in video footage using computer vision is one such area. Although human detection is somewhat primitive, using that data to produce various results like recognizing postures, predicting behaviors, predicting paths are very advanced fields and they have very much room left to grow. Various algorithms, approaches are available today to accomplish the above kind of tasks, from classical machine learning , neural networks to statistical approaches like Bayes theorem, Hidden Markov Models, Time series, etc. This paper aims at exploring algorithms and various approaches that have been used by researchers in various scenarios that are related to post processing the data taken from video footages by detecting and analyzing human figures in them.

Are you a cricket fan? Well, most of the time the answer would be “yes”. But are you satisfied with the current situation of the Sri Lankan cricket team? You are very frustrated. Isn’t it? What do you reckon? Doesn't this happen because our team is unable to select the best eleven players? You don’t need to worry anymore. We found a better solution to select the best team. That is a Data mining and machine learning system for selecting a winning cricket team. We went through four different types of analysis as individual player performance, combined player performance, frequent player performance and finally the outcome prediction for a given cricket match. For outcome prediction of a given match we used our individual performance details and we got 70% accuracy on that.