Proceedings of the 12th Annual Sessions, Sri Lanka Association for Artificial Intelligence
Artificial Intelligence in New Media and Entertainment
Prof. R P C Janaka Rajapakse
Abstract– Artificial Intelligence (AI) technology is revolutionizing the development process. Learn how major intelligence innovation platforms currently available from deep learning can help you solve major design/development problems and bring real value to your business. Architecture Engineering & Construction, Manufacturing, and Automotive are just some of the industries experiencing the most exciting shift in the design process in over a decade.
And the graphic hardware programing has proven to be effective at solving some of the most complex problems in the entertainment industry. It started out as an engine for simulating human imagination which produces up the realistic virtual worlds of video games and Hollywood films. Today, virtual reality (VR) is the latest hype for game developers for engaging audiences from content consumption perspective. However, the AI technology has much broader applications. As results of the latest advancements in AI technologies are becoming approachable for engineers, architects and visualization specialists. Meanwhile VR/AR and 3D printing are enabling new workflows in design, prototyping, visualizing and marketing products. Are you ready for this technology explosion? What tools and AI-ready technology can amuse you and your employees get entertainment and safely? This speech will focus on different available virtual-reality (VR) and 3D printing technology, with the purpose of implementing AI into their workflows. We will have an in-depth discussion both about our developments, as well as about future perspectives.
AI in Your Pocket: Myth vs Reality
Dr. Ruwan Weerasinghe
Abstract-The once spurned technology ‘artificial intelligence’ has made a comeback! And with a vengeance! AI is now the darling of not just some techies, but mainstream CXOs! What caused this dramatic shift and how realistic is the expectation that society now bestows on it?
This talk will explore some of the reasons for the current resurgence of interest in AI, and try to disentangle the hype from the reality ‘on the ground’. It will also open up a conversation on what as AI professionals, our own opinion on this should be, as we encounter it at business meetings and the very corridors of power.
Ontology Design and Development for University Admission in Sri Lankan Universities
T S D H S Wickramarathna, C Anutariya
Abstract-University Admission process is a complex and time-consuming task which requires experts’ advice in all the stages. Since the different stages belong to different organizations, it is rare to find experts having an overall knowledge of the process, from whom anyone can get a complete guide. Local universities in Sri Lanka are fully government funded but able to admit only around 10% of the qualified students. This leads to a high competition in admission but proper guidance in the admission process is still not established. A research has been conducted to develop a recommendation system to encapsulate the complexity of university admission process and help students to take the most suitable decision. In this paper, design and development of the Ontology, which is the knowledge base for the recommendation system is discussed. This Ontology model the undergraduate admission scenario in Sri Lanka, together with the career path mapping, so that students can far see their future opportunities before taking decisions on advanced level subject selection and university application. Ontology was developed using Protégé and was tested by running SPARQL queries. SPARQL query results confirmed that the ontology design is comprised with the real world.
Mind Uploading with BCI Technology
S Welikala, A S Karunananda
Abstract- Mind uploading has been an emerging field of Artificial Intelligence. This area has been primarily influenced by the research and developments in Brain Computer Interfacing (BCI). This paper presents a research into BCI and its influence on mind uploading. In this context first, a description of the essentials of brain waves, BCI technology, and analysis and processing of brain waves are presented. Secondly, it proceeds to discuss the mind uploading technology with reference to EEG based brain waves analysis through BCI. The research has revealed the recent developments in mind uploading, its applications such as data backup, immortality and uploaded astronauts. More importantly, it has discovered legal and social concerns of mind uploading as a matter, which has not been given adequate attention, despite the numerous developments in the technological landscape of mind uploading research
Deep Learning Techniques for Speech Recognition and Audio Transcription for Content Captioning
D G J B Wimalarathne, S C M de S Sirisuriya
Abstract-Speech recognition has taken a tremendous interest in the field of Natural Language Processing. Deep Learning has been playing an increasingly large role in speech recognition and one of the most exciting things about this field is that speech recognition is at a place right now where it is becoming good enough to enable exiting applications that end up in the hands of users. Studies have shown that efficiency of devices increases 3 times, with the availably of speech recognition on daily driving devices like mobile phones, cars etc. Content captioning or generating subtitles is very useful and important since the use of videos for communication have been grown phenomenally in the past few years. However, for this, it still need human influence to do a good captioning. But, it is possible to do this with a better efficiency and accuracy with deep learning. This paper provides an overview of how traditional ASR pipeline process speech transcription and how deep learning techniques used to replace modules in traditional ASR pipeline. Finally, describes the implemented speech recognition neural network model built on Tensor flow platform. This model is capable of recognizing English spoken digits from 0 to 9 and transcript it to text. Currently model has achieved around 70.19% accuracy over different training and testing data. This prototype model is reusable for any amount of words in any language with the availability of more data and computing power.
Dictionary-Based Machine Translation System for Pali to Sinhala
R M M Shalini, B Hettige
Abstract-Machine translation systems are language translation tools that are also capable to be used as language learning tools. This paper presents a machine translation system that has been developed as a language learning tool for the Pali to Sinhala. This Pali to Sinhala translation tool can translate simple Pali sentences into Sinhala through the dictionary-based approach. The translation system comprises of three modules, namely the Pali morphological analyzer, the dictionary-based translator, and the Sinhala morphological generator. The Pali morphological analyzer uses an affix-spiriting approach to identify a relevant Pali root word for the existing Pali word. The Pali to Sinhala translator identifies an available Sinhala-based word for the existing Pali-based word with the support of the Pali-Sinhala dictionary. The Sinhala morphological generator generates appropriate Sinhala words by using the Sinhala root word and the relevant morphological information of the Pali word. This translator uses word-level translation and gives attention only to source and target morphology. The Pali Sinhala translator has been successfully tested for simple Pali sentences as a language learning tool for the grade 6-9 Dhamma schools
On Optimizing Deep Convolutional Neural Networks by Evolutionary Computing
M U B Dias, D D N De Silva, S Fernando
Abstract-Optimization for deep networks is currently a very active area of research. As neural networks become deeper, the ability in manually optimizing the network becomes harder. Mini-batch normalization, identification of effective respective fields, momentum updates, introduction of residual blocks, learning rate adoption, etc. have been proposed to speed up the rate of convergent in manual training process while keeping the higher accuracy level. However, the problem of finding optimal topological structure for a given problem is becoming a challenging task need to be addressed immediately. Few researchers have attempted to optimize the network structure using evolutionary computing approaches. Among them, few have successfully evolved networks with reinforcement learning and long-short-term memory. A very few has applied evolutionary programming into deep convolution neural networks. These attempts are mainly evolved the network structure and then subsequently optimized the hyper-parameters of the network. However, a mechanism to evolve the deep network structure under the techniques currently being practiced in manual process is still absent. Incorporation of such techniques into chromosomes level of evolutionary computing, certainly can take us to better topological deep structures. The paper concludes by identifying the gap between evolutionary based deep neural networks and deep neural networks. Further, it proposes some insights for optimizing deep neural networks using evolutionary computing techniques
Clustering-based Augmented Tabu Search for Energy Minimizing Vehicle Routing Problem (EMVRP)
G W H H P Rathnakumara, T D Rupasinghe
Abstract-Application of metaheuristics for Energy Minimizing Vehicle Routing Problem (EMVRP) has become extremely important because of practical relevance. The EMVRP aims at serving dynamic demanding customers distributed throughout the world while consuming minimum possible energy, which represent as a product of load and distance. The NP-hard problem nature urges the researchers to utilize metaheuristics to solve them in non-polynomial time. Although there exists several metaheuristics, as of to date the EMVRP is only formulated and solved using the Genetic Algorithm (GA). According to the authors’ knowledge there is only one instance reported in literature of metaheuristic based on GA on EMVRP. There are no studies reported on the development of TS based metaheuristic for the EMVRP and use of machine learning to guide the local search in formulating EMVRP. In this study, at first the authors formulate EMVRP using Tabu Search (TS) and evaluate its performance with test cases from the CVRPLib repository for the vehicle routing problems (VRPs). As the second stage, the authors improve the local search in the data set embedding machine learning (ML) techniques in the developed TS algorithm and critically evaluate the applicability using the same test cases. The study introduces a new TS based formulation for EMVRP for the literature as well as derives the fact of augmenting ML brings super performance in metaheuristics in solving EMVRP. Furthermore, the study lays foundation in filling the knowledge gaps in the current literature and proposes future research directions in machine learning augmented metaheuristics to enhance solving the EMVRP
Predicting Floods in North Central Province of Sri Lanka using Machine Learning and Data Mining Methods
H Thilakarathne, K Premachandra
Abstract-The frequency of the occurrence of natural disasters at present has increased due to changes in global and regional climate. Hence, being able to forecast natural disasters has shown to be extremely useful in mitigating the loss of damages to the mankind. In this study, a hybrid model is developed for predicting the occurrence of floods in the North Central Province of Sri Lanka, using machine-learning techniques with artificial neural networks. The hybrid model developed combines two sub predictive models. First model predicts the future weather-related measurements using time series modeling. The second model, which is a binary classification machine-learning algorithm, predicts the probability of the occurrence of flood incidences in a future month using the forecasted weather values and historical flood data. The results show that all probabilities predicted by the hybrid model are a 91.7% match to the actual flood occurrences. Hence, this model could be adopted to predict flood occurrences of any region in Sri Lanka using historical weather and flood related data of the region of interest. The predictive model developed has been published as an Application Programmable Interface on Microsoft Azure cloud, illustrating the practical usage and feasibility of machine learning techniques in developing modern intelligent applications.
A Genetic Algorithmic Approach for Optimizing Supply Chain Network Designs for Retail Distribution Networks
I U Munasinghe, T D Rupasinghe
Abstract-The modern industries are compelled to become more competitive for providing the quality product at a minimum cost. In particular, companies have to analyze the efficiency of their logistic operations and need to adopt a holistic view of the entire supply chain network. The network design optimization attempt to optimize the flows of goods traverse among the main entities of the supply chain, not limited to the suppliers, warehouses, distribution centers, retailers and customers. In this study, Network Optimization (NO) will be carried from the perspective of the Retail Supply Chain (RSC) and study proposes a Genetic Algorithm (GA) for optimizing a novel mathematical model of the retail SCND. Most of the real world scenarios involve multi-source, multi-stage and multi-product with large instances. Therefore, the developed GA is more beneficial for real world scenarios strategic decisions such as location and allocation decisions and also to perform powerful ‘what if’ analyses. The authors have elaborated the GA with numerical examples and compared the results. The outcome of the study proposes the GA approach to reduce the distribution network design cost and the most appropriate number of facilities to minimize total supply chain costs to match the organization’s service goals
Knowledge Discovery with Data Mining for Predicting Students’ Success Factors in Tertiary Education System in Sri Lanka
K T S Kasthuriarachchi, S R Liyanage
Abstract-Knowledge discovery in educational data would be so basic to determine better expectations on the undergraduates. Distinguishing proof of the components influence to the execution of undergraduates in light of various attributes will be supportive for instructors, educators and managers viewpoints. This paper endeavors to utilize different data mining ways to deal with find forecast manages in undergraduates’ data to distinguish the components influence to the scholarly accomplishment in their tertiary education. The approach of this exploration observed the aftereffects of three mining algorithms with about 3800 undergraduates’ records and the calculation which demonstrated the most elevated exactness has chosen as the best model and the connections acquired through that were gotten to foresee various elements against the objective of whether they will get the degree or not following three years of the university life. Naïve Bayes, Decision Tree and Support Vector Machine were used in predicting the most affecting factors to the performance of students. According to the prediction accuracy levels, the results of Decision Tree were selected since it outperforms the rest for the selected data set. Finally, the results were evaluated using a correlation analysis to select the most prominent factor. According to the test, the age, past failure modules, performance of past semesters were selected as the most influencing factors to the success or failure of the students in tertiary education system in Sri Lanka
A Simulation-based Analytical Approach to Enhance Distribution Networks in Pharmaceutical Supply Chains
Q J D N N Rathnasiri, T D Rupasinghe
Abstract-This paper focuses on the most efficient factors to be considered when designing a distribution network for pharmaceutical supply chains (PSC). When considering the impact on the consumers, distributing pharmaceutical products has become a crucial process than distributing other commercial goods. The study has a two-fold approach. Where in the first phase, the paper represents a comprehensive systematic review of literature based on PSC in terms of simulation and modelling techniques and the distribution networks. In the second phase, it designs distribution network models to simulate pre-identified clusters of the pharmaceutical based on the inherent characteristics. Finally, the most efficient distribution network designs are identified for each cluster. Here, four different distribution network designs were modelled using SupplychainGuru® simulation and modelling platform. These models help make strategic decisions such as designing and operating different distribution networks to better suit the product characteristics of a given PSC by contributing to reduce overall supply chain cost, reduce overall supply chain time, increase product availability and to increase product quality in order to deliver better quality products to the patients in need.