Prof. Cyrus Shahabi, Director of the Integrated Media Systems Center (IMSC), made a significant contribution to this year’s International Conference on Machine Learning (ICML), one of the premier conferences in the field of machine learning. Prof. Shahabi delivered an oral presentation on his paper titled, Theoretical Analysis of Learned Database Operations under Distribution Shift through Distribution Learnability.

ICML 2024 was highly competitive, with 9,473 total submissions from researchers around the globe. Of these, only 2,609 papers were accepted for publication, and a mere 144 were selected for the coveted oral presentation slots, underscoring the high regard for Prof. Shahabi’s work.

Prof. Shahabi’s presentation highlighted the work of his PhD student, Sepanta Zeighami, now a postdoctoral researcher at UC Berkeley. Their research focuses on using machine learning for database operations such as indexing, cardinality estimation, and sorting, which have shown significant performance benefits. However, when datasets change and data distribution shifts, learned models can suffer from performance degradation, sometimes falling below non-learned alternatives. The paper addresses this issue by presenting the first known theoretical framework that characterizes the performance of learned models in dynamic datasets. Their analysis introduces the concept of distribution learnability, offering theoretical bounds that explain when and why learned models can outperform traditional methods, providing crucial insights for their practical applicability in real-world systems. See below for the Viterbi’s news on this work: https://viterbischool.usc.edu/news/2024/07/usc-at-the-international-conferences-on-machine-learning-icml/

The ICML conference, held annually, is a key platform for the dissemination and discussion of groundbreaking research in machine learning. This year’s event, featuring leading academics and industry experts, provided an opportunity for researchers to share insights and advancements that will shape the future of the field.

Prof. Shahabi’s selection for an oral presentation at such a prestigious conference highlights the importance and innovative nature of his research, further solidifying his position as a leading figure in the machine learning community.