- Project Year:
2025
- REU Student (s):
Winston Li | Rutgers University-New Brunswick NJ
- Student 1 Institution:
Rutgers University-New Brunswick
- Project Mentor:
Periklis Papakonstantinou
- Project Mentor Area:
Management Science and Information Systems
- Project Abstract:
Verifiable sources of randomness are a ubiquitous resource for modern algorithms, like machine learning and cryptography. Existing methods like distribution and statistical testing can be used to check the randomness of a source. While the distribution testing is more theoretically sound, statistical tests are often the only computationally feasible choice. This report analyzes ways to bridge techniques from both fields, reframing the NIST test suite in the language of distribution testing and using randomness extractors to empirically validate these tests. This includes technical optimizations needed to complete the experiments in a reasonable amount of time. Additionally, we explored using compression as a measure of randomness, although our results demonstrate that it is unlikely to be stronger than statistical tests. More importantly, the framework used to test the compression algorithms involve notions of epsilon-closeness, which could be used to design more robust uniformity tests.
