Andrew McGregor
University of Massachusetts, MA·Computer Science
Many of today's datasets are too large to store, so computers must analyze them as a stream: each item is seen once, in a small amount of memory, and summarized on the fly. Data stream algorithms do this with proven limits on running time, memory and accuracy.
- Fields
Computer Science
Mathematics & Statistics
- Topics
data stream algorithms
streaming algorithms
adversarial robustness
- Grants
- 1 known grant, through Jun 2027
What they work on
- Develop adversarially robust data stream algorithms that stay accurate when inputs are chosen adaptively in response to the algorithm's earlier outputs, as happens in interactive data analysis.
- Design parameter-free and non-adaptive streaming algorithms that need no prior knowledge of problem-specific parameters, avoiding the costly approach of running multiple copies with different guesses.
- Determine when randomized streaming methods can be matched or approximated by pseudo-deterministic or deterministic algorithms, which give reproducible outputs for experimental science.
- Create fault-tolerant data stream algorithms that guarantee correct results despite hardware or communication errors.
More about the research
Those guarantees, however, often depend on hidden assumptions about how the algorithm will be used, and many algorithms stop working when the assumptions fail. This project aims to design streaming algorithms that keep working when those assumptions break.
Why it matters
Fields such as finance, healthcare, genomics, high-energy physics, cloud computing and machine-learning training all process huge data streams. Streaming algorithms that resist adaptively chosen inputs, need no tuned parameters, give reproducible answers and tolerate hardware faults could make large-scale data analysis more reliable for these users.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Vinodchandran Variyam
University of Nebraska-Lincoln, NE·Computer Science
Initiate the study of a "right to forget" data stream model, in which items can be required to be deleted, motivated by privacy concerns and the growth of data generation.
Michael Bender
Stony Brook University, NY·Computer Science
Design efficient methods to build graph-based vector search indexes such as HNSW and DiskANN, reducing their currently very high construction costs.
Minlan Yu
Harvard University, MA·Computer Science
Design a stateful network switch architecture that can manage, compute on and share application state inside the network, enabling in-network computing for distributed and high-performance computing applications.
The school
University of Massachusetts: 28 labs on PI Pages, $282M a year on research, 337 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.