Proceedings of the 11th Annual Sessions, Sri Lanka Association for Artificial Intelligence

(International Conference on Emerging Trends in Artificial Intelligence- ICETAI 2015)

29th December 2015 – The Open University, Colombo


Towards Intelligent Sensor Fusion based Visually Impaired Navigation: An Assistive Technology Framework

S. Silva , S. P. Wimalaratne

Abstract– This paper presents the development of an electronic navigation framework for blind and visually impaired persons. This proposed approach aims to intelligently fuse the surrounding information sensed via ultrasonic sensors and vision sensors. The intelligent component of the prototype developed would serve in several facets including object recognition and computational performance optimization within the embedded system software. The prototype developed for field testing in indoor as well as outdoor environments would be used as assistive technology for visually impaired. The wearable device developed would provide the feedback via tactile cues. The current status of the research and the future developments are presented in this work.



Multi-Agent Based Train Controlling System

A. P. Kasthoori, K. de Soyza, J. G. V. Priyanka

Abstract– Application of agent based system is very important in fast moving world with modern technologies in order to deliver maximum productivity using minimum resources. In the railway domain, this will account for reducing unnecessary train traffic on rail network and reduces operational overhead. This research paper discusses how this technology has been applied for the benefit of Sri-Lanka Railways.



Modeling the Learning Ability of Fish in an Artificial Fish Simulation using Reinforcement Learning

G. M. T. C. Galahena, S. P. Wimalaratne

Abstract– Fish display a considerable amount of learning skills in activities like foraging and defense (ex: locations and quality food patches, areas where certain predators are in, etc.). But most of the existing models of fish behavior do not simulate the learning patterns of fish. It makes those models less realistic. This research tries to fill that gap by creating a model with learning ability which is more similar to the actual behavior of the fish. The main focus of this research will be on the learning involving in foraging and defense of the fish. The system will be a comprised of multiple agents to represent fish and each agent will act individually. The senses and the locomotion abilities of the agents in the simulation will be generalized representations of actual fish. And the learning will be simulated using machine learning algorithms.



Intelligent Controller for Inverted Pendulum

P.M.S.D.K Pathiraja, G Anthonys, H.D.N.S Priyankara

Proportional Integral Derivative (PID) controllers are widely using in many control applications, but modeling and tuning of PID controllers are not easy tasks. It needs mathematical modeling and previous experience for doing it. In modern control applications, intelligent control methodologies are used in various approaches. This study focuses on such type of an intelligent control application, inverted pendulum control using an intelligent controller. Inverted pendulum is a famous classical controlling problem. It is unstable and making it stable is a hilarious task. Development of an intelligent controller for feedback’s digital pendulum plant is present in this paper. This plant has a single input and two outputs. The hybrid Neuro-Fuzzy architecture used in this paper is ANFIS (Adaptive Neuro-Fuzzy Inference system). This architecture has the essence of the intelligent controllers, neural networks and fuzzy inference systems. Seven linguistic variables are used for each output, Pendulum angle and cart position as well as another seven linguistic variables used for only input control voltage of the motor. To train the neural network a hybrid training algorithm is used. ANFIS controller is showing better result in settling angle, stability, less overshoot and less settling time when compare with existing PID controller.



Ontology Development for Sri Lankan Medicinal Plants: A Knowledge Representation

P. Jeyananthan, E. Y. A. Charles, D. A. S. Atukorale

Our ancestors consumed medicinal plants through their day to day activities which leads them to a long and healthy life. It plays an important role in their life. But in this fast moving world the value of these plants are gradually going down. As they take comparably long time to give relief from the disease, the people are not interested in these medicines. Fast and modern foods are replaced our traditional food system which lessen the intake of herbs through foods. We have responsibility to aware our future generation about this medicinal plants and their usage, because they are in danger of extinct. The knowledge about these plants are scattered over the web and some experts are still here to carry this knowledge to our future. This research aims to collect all these knowledge from different places and build a knowledge base system for Sri Lankan medicinal plants. This knowledge base consist plants parts, diseases, preparation method and mixtures if any. Ontology is the technique used to develop the knowledge base. To develop the ontology, Protégé which is an open source is used. SPARQL is the language used to query over this knowledge base.



Agent-based Solution for Improving Abstracts
A. M. T. B. Adhikari, A. S. Karunananda

Abstract-Writing abstracts in a comprehensive and meaningful manner is a challenge for any researcher. However, an abstract includes selected set of verbs, standard phrases and other good practices of structuring the contents. A research has been conducted to develop an automated solution for improving abstracts. This solution is based on multi agent systems technology and natural language processing together with ontology of commonly used verb phrases and other good practices. The system has been developed with nine agents, namely, coordination agent, parser agent, problem agent, solution agent, conclusion agent, content agent, synonym agent, improvement agent and restructure agent. These agents receive an abstract and interact with each other to reach consensus on the possible improvements to the abstract. For instance, problem agent and solution agent may agree on the proportion of respective contents within the abstract. The Stanford CoreNLP Natural Language Processing Toolkit has been used to develop parser and JADE has been used for development of the entire multi agent system. The system has been developed with JAVA to run on Windows. It has been incrementally tested, and shown interesting results related to improving abstracts.



Question Matching Technique to Find Answers

P.P.G Dinesh Asanka, A. S. Karunananda

Abstract-In today’s business world, there are lots of knowledge systems and users need to find answers from existing knowledge bases. Due to the complexity of the systems and fat contents of them, it is difficult for users to find appropriate answers to their questions. Knowledge bases can be configured to mapped answers to questions. When end user enters a question, using natural language processing and text mining techniques, the question is matched with the existing question in the question bank in the knowledge base and matched answer and any other related contents (if exists) are provided to the end user. Proposed technique was evaluated with Sri Lankan 1978 constitution knowledge base and it was found that many questions can be matched with higher accuracy.