Modern web search, recommendation systems, and generative AI applications rely heavily on vector search, also known as Approximate Nearest Neighbor Search (ANNS), to retrieve semantically related objects based on the geometry of their learned representations. As these applications often involve collections with billions or even trillions of items, the design of efficient and scalable ANNS algorithms is essential. In this talk, I will trace the evolution of ideas from small-world phenomena to the algorithms that now underpin large-scale search and retrieval. I will highlight the core principles behind these methods, their practical challenges at web scale, and their role in enabling today’s search and generative AI systems.
Ravishankar Krishnaswamy is a principal researcher at Microsoft Research India. He completed his PhD from Carnegie Mellon University in 2012, and was a Simons Postdoctoral Fellow at the CS Department in Princeton University from 2012-2014. His recent research has been on designing new algorithms and data structures for very large-scale vector search. He is also interested in approximation algorithms, especially for graph-(connectivity/flow) problems and clustering problems, and in studying models which incorporate uncertainty in the input such as online algorithms and stochastic optimization.