Towards Practical Distribution Testing

Explore the challenges and advancements in distribution testing for high-dimensional samplers in this 23-minute lecture from the Workshop on Local Algorithms (WoLA). Delve into the need for verifying statistical correctness in safety-critical software systems employing samplers. Examine the intractability of black-box testing in high-dimensional settings and discover the potential of "grey-box" models like conditional sampling as promising alternatives. Learn about the development of grey-box algorithms that are efficient both in theory and practice, with a focus on the first polynomial query algorithm for TV distance estimation in the conditional sampling model. Gain insights from Yash Pote of the National University of Singapore as he presents this work towards practical distribution testing.