Oren Netzer

This talk will cover the fascinating world of coresets and how they can be used to decrease model development time, reduce training time and improve model quality. A coreset is a methodology from computational geometry for sampling a dataset without losing any important information or outliers. Coresets are more reliable than other forms of sampling and can serve as a replacement of the full dataset for the purposes of data exploration, feature engineering, model training and model tuning. In this talk, we will briefly explain about the theory behind coresets, show how to easily build and manage coresets and how to easily add them to your existing pipeline using our proprietary, commercial Python library. Finally, we will provide a few examples of the superior results that can be obtained by using coresets and compare them to other methods.