Kernel density estimation techniques and their applications in ML


Event details

Date 05.07.2024
Hour 14:0016:00
Speaker Ekaterina Kochetkova
Category Conferences - Seminars
EDIC candidacy exam
Exam president: Prof. Mika Göös
Thesis advisor: Prof. Michael Kapralov
Co-examiner: Prof. Ola Svensson

The Kernel Density Estimation (KDE) problem finds applications in many practical tasks involving data analysis and approximate fast computations. Naturally, it has become important to machine learning, and attracted great research interest. We present two papers showing how different techniques from this line of work can be applied to obtain practical tools for training large language models, as well as one recent paper providing a result on a problem used as a building block in a particular approach to the KDE problem.

Background papers
  1. Janardhan Kulkarni, Victor Reis, Thomas Rothvoss. "Optimal Online Discrepancy Minimization."
  2. Moses Charikar, Beidi Chen, Christopher Re, Erik Waingarten. "Fast Algorithms for a New Relaxation of Optimal Transport". Proceedings of Machine Learning Research vol 195:1–32, 2023.
  3. Josh Alman, Zhao Song. "Fast Attention Requires Bounded Entries". 37th Conference on Neural Information Processing Systems (NeurIPS 2023).

Practical information

  • General public
  • Free


EDIC candidacy exam