Sketch it: continuous parallel/distributed analysis with controllable accuracy and CPS/HW awareness

Potential supervisors: 

Motivation: Sketches are data objects supporting continuous/streaming analysis and summarization (#uniques, frequencies, peaks, top-X, percentiles, distributions, KPIs) through updates and queries:  commonly employed in Streaming Data Analysis and Machine Learning e.g. by Google, IBM research, FB, MS research; necessary for supporting in-memory analysis of high-rate data; common need in cyber-physicals systems and applications such as Compressed Sensing, Networking, Security. Their efficiency in parallel and distributed algorithmic implementations is highly interesting, to enable new applications and services in e.g. automotive industry.



Challenges and Possibilities: This thesis is about studying sketch/synopsis streaming analysis and parallel algorithmic implementations with varying parameters, including algorithmic ones and application/platform-dependent ones. Metrics of interest include efficiency in time, accuracy, scalability, studied through empirical evaluation and analytical argumentation where possible. During the thesis, you will get familiar with methods in the literature including the group’s contributions, you will be able to build on existing open-source implementations (with possibilities for new ones) and test and compare results with relevant applications. (Configurable for 1-2 students)

Date range: 
October, 2021 to October, 2024