The Amrita Multidimensional Data Analytics Lab has been set up to enable research in large scale intelligent information systems. With the world facing unprecedented challenges, technology has a major role to play in efficiently solving many of them. Our goal is to develop large-scale information systems that can help tackle challenges in education, healthcare and support smart buildings and cities. Towards this we focus on end-end solutions that include designing innovative data models, scalable data management solutions, novel indexing techniques, intelligent retrieval and event-detection algorithms.
Faculty Incharge of Lab: Dr. Vidhya Balasubramanian
PI – Dr. Vidhya Balasubramanian
Co-PIs : Dr. C.K.Shyamala and Dr. G.Jeyakumar
Funding Agency: DST – NRDMS
Amount: 46.1 Lakhs
Asset and people tracking solutions are crucial in providing seamless operations devoid of clashes between services and elements, especially when designing smart environments for large hospitals. Here monitoring mobile patients, locating important personnel, detecting and locating crowds, locating and tracking essential surgical assets and verification of access control restrictions with respect to people and assets, are important functionalities to be provided by the smart space. In this project, we aim to achieve the above by developing a low-cost, end-to-end system that can help configure and operate a smart hospital, which supports the above functionalities. The proposed system includes a sensor infrastructure, smart algorithms for asset and people tracking applications over this infrastructure and the underlying database system to support these applications. Research focus includes development of effective data management and representation of real-time streams of moving objects and sensor data, cost effective localization, ambient sensing and intelligence, and distributed inferencing.
Products(Copyright (c) 2018 Amrita Vishwa Vidhyapeetham, Dr.Vidhya Balasubramanian)
PI : Dr.Vidhya Balasubramanian; Co-PI: Dr.Latha Parameswaran
Funding Agency: DST – NRDMS
Amount: 51.3 Lakhs
An indoor information system is a system that gives information about the details of a building such as the different spaces within the building, exits, and their properties and also about other entities in the buildings like utilities, objects like tables, shelves etc. It not only allows the users to view the building layout but also allows end-users to query the system for information like exit ways, paths, location of rooms etc. Existing indoor databases are primarily developed for specific applications like navigation systems, 3D visualization of buildings etc. The goal of this project is to create a hierarchical extendible data model for indoor information which can represent buildings, entities within it, utilities, objects inside the building, and sensors inside the building for object tracking and localization, so that the represented data is easily stored and queried using a spatio-temporal database. An indoor information system is being developed as part of this project which can support indoor asset tracking and real-time navigation over this data model. In addition novel indoor localization algorithms are being developed for people tracking and asset tracking. These algorithms aim to use cost efficient technologies and are designed to improve the accuracy of indoor people and asset tracking applications. The project was funded by NRDMS, DST, India (NRDMS/11/1925/012) and was completed in December 2015.
Sanctioned by ICPS Division of DST for Rs. 40 Lakhs
The goal of the proposed project is to develop an efficient Cyber-Physical System for providing solutions for sustainable water management. This project aims to develop solutions and a testbed for monitoring water usage and leaks, and improve efficiency of waste water treatment plants. Towards this we integrate intelligent algorithms and modeling from the cyber perspective, efficient energy management, novel wireless communication node design and placement for the development of intelligent sensing and actuation techniques. Development of novel water network models and simulations to support development of efficient algorithms for leak detection, and a prediction of spread of contaminants is a major thrust in this project. The testbed comprises of sensing and communication nodes at different points in the system for monitoring water usage and leaks and quality of the recycled water. The data from different nodes will be processed in a distributed manner and this processing is designed to be efficient. Intelligent data analytics and decision algorithms will be employed to derive actuation signals which are wirelessly transferred to the hardware controllers.
Indoor Localization depends heavily on the infrastructure like WiFi, Cellular Towers, BLE beacons etc. This project comprises of methods to support indoor localization using the inbuilt smartphone sensors and communication capabilities using a cooperative manner without need for external infrastructure.
Detecting events, counting people, understanding the density of people inside an environment require ambient intelligent systems. However it is essential that the system is cost effective, privacy preserving while being efficient. In this project we design ambient intelligent systems for people counting, intelligent appointment management and crowd density mapping in indoor spaces using technologies like WiFi, BLE Beacons, RFID, IR and Ultrasonic sensors. Distributed and cooperative intelligence is employed to improve the event detection accuracy.
Indoor pervasive applications depend on re- liable indoor localization solutions. Indoor localization using WiFi is gaining ubiquitous usage owing to its simplicity and inexpensiveness. A conventional method of localization is trilateration, which can be accomplished using signal strength or time of flight of a radio signal between receiver and transmitter. However, trilateration is prone to errors in accuracy that can occur due to various factors. A common reason for the failure of trilateration is due to the errors in distance estimation resulting in a poor quality of trilateration. We develop adaptive non-fingerprinting techniques which adapt the basic trilateration for robust, efficient and calibration-free indoor localization.
The goal of this project is to improve the ease of generating knowledge representations, so that semantic search can be better supported. The project aims to simplify the process of knowledge representation by using statistical approaches and enhancing their expressivity using graph theoretical approaches. The graphs are generated both at semantic level and document level. Using these representation we develop novel information retrieval applications like web service recommendations for GIS web services and scholar article recommendations based on the sequence of recommended reading.
Investigator : Dr. Vidhya Balasubramanian
Asset tracking is an important requirement in indoor pervasive systems. However the current techniques require many RFID readers, and expensive fingerprinting to get reasonable tracking accuracy. We develop novel calibration free algorithms as part of this project for efficient RFID based asset tracking solutions. One such solution, the InPLaCE RFID system estimates the position of an object within a 3d clutter by employing a robust translation model that accounts for the properties of the clutter and helps compensate for estimation errors over existing path loss models. Simple bilateration can be used using this path-loss model to accurately locate objects. We also developed a novel hybrid indoor localization algorithm that combines the transmission power level control and the signal strength information in an intelligent manner to locate assets accurately. The algorithm does not require much calibration and is easily scalable. Additionally, it allows for both coarse and ne-grained location estimation depending on application requirements.
Current product recommender systems only allow users to select from off the shelf products. However if they wish to mix and match components to create customized products, current recommendation systems do not support that. Our project expands the way recommendation systems work so that they can suggest the products along with their customizable components to customers and give them a greater choice in shopping. As a part of this project new models to represent this problem and novel and efficient solutions for the same are being developed. Novel recommender systems for scholarly articles, which can help with a scholar organize his/her research and get recommendations on what to read next is
Funded by: This is funded as a part of a larger MHRD Project, E-Learning Research Center, Amritapuri.
This project is a lecture browser system that helps users search a large corpus of lecture videos efficiently. This system helps the user to search the video both at the video level and segment level. A prototype system has been developed and has been hosted on our servers for testing. In addition better metadata extraction approaches have been developed. A tool has been developed based on TESSERACT and GOCR that extracts text from slides that are embedded in the lecture video. Using the extracted content from the audio and video modalities an effective approach for keyphrase extraction and lecture segmentation has been developed. The keyphrases and segments generated by this approach help summarize the content of the lecture videos and aid in better content based search and retrieval.
Investigator : Dr. Vidhya Balasubramanian
ACIST - Keyphrase Extraction
ACIST - Knowledge Representations
Statistical Semantic Networks for Concept and Document Relatedness