07/10/2023
By Niket Kathiriya
The Richard A. Miner School of Computer & Information Sciences invites you to attend a master’s thesis defense by Niket Kathiriya on "Iterative Forgetting: A novel online data stream regression method."
Candidate: Niket Kathiriya
Degree: Master’s
Defense Date: Wednesday, July 19, 2023
Time: 1 to 2 p.m. EST
Location: Dandeneau 309, North Campus, and via Zoom.
Thesis/Dissertation Title: Iterative Forgetting: A novel online data stream regression method
Committee Members:
Kshitij Jerath(advisor), Mechanical Engineering Department, University of Massachusetts Lowell
Cindy Chen, Computer Science Department, University of Massachusetts Lowell
Tingjian Ge, Computer Science Department, University of Massachusetts Lowell
Abstract:
In today's interconnected world, time-sensitive systems are essential across various domains, demanding ultra-low latency for real-time decision-making and accurate predictions. These systems generate continuous data streams characterized by unbounded size, high arrival rates, varying velocity, and possible concept drift. Traditional regression techniques often struggle to keep up with the velocity and volume of data streams, necessitating innovative approaches capable of handling dynamic data while ensuring timely predictions. This research focuses on developing a fast data stream regression model that balances speed and accuracy efficiently, meeting the stringent demands of time-sensitive applications. While the proposed model may compromise marginally on accuracy compared to state-of-the-art methods, its notable strength lies in its significant speed advantage, making it considerably faster. Furthermore, the approach includes a systematic method for identifying and discarding expired instances in large time-series databases or data streams. The contributions of this thesis include the novel data stream regression approach and an automated method for removing expired instances. Performance criteria such as query response time, estimation accuracy, training time, and model size are considered to ensure fast and accurate responses for time-sensitive systems while handling unbounded data streams.
As the proposed solution facilitates faster decision-making in time-sensitive systems, its utilization could be in various industries such as financial markets (predicting stock prices, exchange rates, etc.), manufactoring and quality control (predicting defect rates and product performance), energy management (predicting demands), IoT and Sensor networks (predicting maintenance and anomaly dectection), and Transportation (predicting traffic flow, travel duration, etc.)