Spatial Network Embedding
Graph (or Network) embedding techniques (a novel form of deep-learning technique for graph-related problems) capture complex latent information of a large attributed network for answering many interesting and useful network analysis tasks. The network embedding essentially learns a low dimensional representation of high dimensional contents of nodes and edges of a given network. These techniques have been recently applied to many real-world problems such as social networks, citation networks, knowledge graphs and so on. In this project, we aim to develop a new form of network embedding particularly focusing in geo-spatial context, and solve many real-life problems like house price predictions, POI recommendations, etc.
Processing Big Trajectory
With the widespread use of GPS-equipped mobile devices and the popularity of mapping services, an unprecedented amount of trajectory data is becoming available for data analytics applications. These trajectories are captured in the form of a sequence of time-stamped locations for moving objects’ (e.g., cars, users,) traces. In this project, we develop techniques for big trajectory data to support a new set of applications that include contact tracing for COVID-19, popular routes for traveling, designing routes for auto-nomous vehicles, etc.
Bangla NLP and Deep Learning
Approximately 228 million native speakers and another 37 million as second language speakers use Bengali as medium of their communications. Unfortunately, compared to other prominent languages such as English, Chinese or German, a little effort has been given in the domain of Bangla NLP. In this project, we tackle several application domains of Bangla NLP that include: Spell Checking and Auto-Correction, Bangla Chatbot, Bengali Literature Analysis, and so on.
Groups in Geo-Social Networks
Various social network sites now allow users to capture their locations through GPS-enabled devices and share them through check-ins or mentions in posts. As a result, socio spatial networks are emerging where each user is associated with a physical location along with the connectivity with other members of the network. In this project we aim to find interesting sets of user groups and subgroups based on different social, spatial, and textual constraints/preferences, which enable business operators to target effective campaigning, friends and families to plan their gathering, etc.
3D City and Real-life Queries
The 3D models of real-life urban structures are becoming widely available through the popular mapping services, such as Google Maps, Google Earth, and OpenStreetMap. Such availability enables us to address many practical applications that require visibility computation in the presence of 3D obstacles. For example, (i) a tourist may wish to find a location to enjoy the best view of a tourist attraction, or a hotel with the best view of the city skyline; (ii) a security company may want to find the suitable positions to place surveillance cameras in a city; (iii) an advertisement company may want to check the visibility of a billboard from the surrounding areas to decide on the billboard’s position and suitable font size. In this project, we cover a wide range of topics that include generating 3D city models from drone images, creating 3D layout of indoor spaces, processing queries on large 3D databases. For more details Video: https://www.youtube.com/watch?v=rcizJtFvQfU, and Codebase: https://github.com/arif-arman/vizq
Personality Traits and Values from Social Networks
Social Networking Sites (SNS) such as Facebook, Twitter, Foursquare and IMDb have become major platforms of communications for users in the web. These SNS allow a user to share ideas, thoughts, and opinions with her friends, family and acquaintances. Every day millions of newsfeeds and tweets are posted in these SNS. The contents of these newsfeeds and tweets provide a rich platform for the researchers to identify cognitive and psychological attributes such as personality, values, and preferences of involved users. We investigate whether these psychological attributes derived from social media interactions can be exploited to support applications like mental health, recommendations, advertisement, and so on.
Applied Machine Learning
Machine learning, in particular deep-learning, has touched almost in every sector of our daily lives, ranging from agriculture to personalized health-advice, from surveillance to crowd-counting. In this project, we propose deep-learning based solutions to critical problems in the context of Bangladesh. Our application areas include, a Bangla Taka recognizer for blind people, a light-weight rice disease recognizer for farmers, and so on.