Research

My current focus of research is in the field of Artificial Intelligence and Natural Language Processing. I have been involved in research activities since 2013 and explored several other related fields also. My research areas include: Artificial Intelligence, Natural Language Processing, Machine Learning, Probabilistic Analysis, Data Mining, Vehicular Network, Routing, and Optimization.

Current research project:

    • May, 2018 - Present: Co reference resolution in Winograd Schemas.
      • Supervisor: Dr. James Delgrande and Dr. Fred Popowich
      • Summary: Concerned about the Turing test’s ability to correctly evaluate if a system exhibits human-like intelligence, the Winograd Schema Challenge (WSC) has been proposed as an alternative. A Winograd Schema consists of a sentence and a question. The answers to the questions are intuitive for humans but are designed to be difficult for machines, as they require various forms of commonsense knowledge about the sentence.

Example: The trophy doesn’t fit in the brown suitcase because it’s too big. What is too big?

        • Answer 0: the trophy
        • Answer 1: the suitcase

Both the candidate referents agrees in number and gender. Therefore, To answer this question, we can not rely on machine learning or any other statistical methods only. The agent needs the following commonsense: "If X does not fit in Y, then X is larger than Y".

In this research, i am trying to combine both NLP and commonsense reasoning techniques to approach the co referencing problem.

List of previous research projects:

    • Jan, 2017 - April, 2017: Optimal routing for bike rebalancing vehicles used in bike sharing systems.
      • Supervisor: Dr Jie Wu.
      • Summary: There are several bicycle rental companies in big cities. Users can rent a bicycle from any dock and return it to any other dock. The result is, at the end of the day, some docs becomes overfull, and some docks become empty. To re-balance the docks, rental companies use special vehicles to carry the bicycles from overfull docks to empty docks. If we consider all the physical constraints, (e.g., gas usage, limited capacity, delay, etc) planning optimal route is a very challenging NP hard problem. We developed an approximate algorithm that allows the vehicles to plan reasonably good route while satisfying all the constraints.
      • Our paper has been published in proceedings of 12th Workshop on Challenged Networks, (ACM CHANTS 2017), Snowbird, Utah, USA.
    • August, 2016 - December, 2016: Optimizing Carpool Scheduling Algorithm for Uber drivers.
      • Supervisor: Dr Jie Wu.
      • Summary: We developed a routing algorithm for Uber drivers to schedule their pickups and drop-offs during carpooling. Our algorithm tries to minimize both service delay and gas usage.
      • Our paper has been accepted to IEEE International Conference on Communications, 2018, Kansas, MO, USA. However, it has not been published yet.
    • September, 2014 - September, 2015: An Efficient Approach of Computing Voronoi Diagram in a Large network
      • Supervisor: Dr. Dr. Mohammed Eunus Ali and Dr. Sarana Nutanong.
      • Summary: In our day to day life, network Voronoi diagram (NVD) of order k has several applications. One of the most useful applications is to find k nearest points of interests (POIs) (i.e., generators) which is applied widely in many aspects of real life. People are often interested in finding k nearest POIs (e.g., restaurants, gas stations, hospitals etc) and choose among them. Construction of NVD in a spatial network (i.e., graph) is a very challenging task as it takes massive computation overhead. However, when the graph is too large to keep in a single computer, the computation becomes even more complex. Real life spatial networks (e.g., road networks) are often very large with millions of vertices and therefore can not be handled in a single computer. For this reason current trend is to use distributed environment where the graph is partitioned and distributed to several computing units. In distributed environment, we have to consider not only the computational complexity but also message passing and synchronization among different computing units (i.e., processes) running in different computers. Several existing models can be used to solve this problem but they does not support enough degree of parallelism. We developed a system that works asynchronously and effectively uses high degree of parallelism to construct NVD of order k in a very large graph. We tested the performance as well as correctness of our model on both synthetic and real graph of USA road network with 11.5 million vertices using gas stations as generators and found a significant improvement compared to existing approaches.
    • January, 2014 - August, 2014: Identifying Spatio-Temporal Crime Pattern (Dhaka Crime Alert!)
      • Supervisor: Dr. Dr. Mohammed Eunus Ali.
      • Summary: Street crime is a prevalent problem in developing countries like Bangladesh. Though this problem has been identified long before, no visible remedy or action can be seen to overcome or combat these street crimes in one of the most populated mega-city, Dhaka, Bangladesh, of the world. In this paper, we propose a novel spatio-temporal street crime prediction model that exploits the historical street crime data of Dhaka city to predict the possibility of a crime in a particular region at a specified time. They key intuition of our spatio-temporal crime prediction model is base on a observation, that is, the location and the time of a future crime follows a strong correlation with the location and the time of past crimes. For example, when a crime happens at a particular place, there is a high probability that the same type of crime will occur again in a nearby place. In our model we capture both space and time proximity of a past crime while predicting a future crime. Experimental evaluation shows that our spatio-temporal prediction model can predict a future crime with sensitivity 79.24%, and specificity 68.2%. As a proof of concept we develop an Android application that alerts a user about the possibility of different crimes in a place at a particular time.
      • Our paper has been published in proceedings of Eighth International Conference on Information and Communication Technologies and Development (ICTD '16) Ann Arbor, MI, USA.
    • February, 2013 - December, 2013: Road Traffic Simulator (DhakaSim)
      • Supervisor: Dr. A. B. M. Alim Al Islam.
      • Summary: Bangladesh, in particular its capital Dhaka, is widely known for traffic jam. Analytical studies on such traffic jam has the potential to lead towards a solution, which can significantly diminish the extent of sufferings resulted from the jam. In this project, we will explore the viability of such analytical study through investigating applicability of currently-used road traffic simulator in the context of Bangladesh. If the currently-used road traffic simulators are not applicable, which may happen due to various phenomena observed over the roads in Bangladesh, we will develop our own simulator. Subsequently, on the basis of our simulation outcomes, we will attempt to lead towards a probable solution for the traffic jam in Dhaka city.
      • We have submitted our paper and poster on MoHCI and our paper has been accepted as well as our demo. Here is our paper link. DhakaSim.pdf