Improved Frequency Estimation Algorithms With and Without Predictions

Improved Frequency Estimation Algorithms With and Without Predictions

Explore frequency estimation in data streams through this 27-minute conference talk by Sandeep Silwal from MIT. Delve into a new algorithm that outperforms classical streaming algorithms in certain parameter regimes without relying on predictions. Learn how augmenting this algorithm with heavy-hitter predictions further reduces error and improves upon state-of-the-art techniques. Discover the empirical superiority of these algorithms compared to prior classical and learned approaches. Gain insights into the joint work with Anders Aamand, Justin Y. Chen, Huy Nguyen, and Ali Vakilian, which will be presented at NeurIPS '23.