AI, Data Science & Cloud Certificates from Google, IBM & Meta — 40% Off
One plan covers every Professional Certificate on Coursera. 40% off your first 3 months — limited time.
Unlock All Certificates
Watch a 40-minute research lecture from the Joint IFML/MPG Symposium at Simons Institute where Kevin Tian from UT Austin presents groundbreaking work on omnipredictors for single-index models (SIMs). Learn about a novel construction that achieves competitive performance on matching losses induced by monotone, Lipschitz link functions when compared against bounded linear predictors. Explore how this new approach requires significantly fewer samples (approximately eps^-4) and operates in nearly-linear time, with further improvements to eps^-2 samples for bi-Lipschitz link functions. Discover the enhanced analysis of the Isotron algorithm in agnostic learning settings, which enables proper learning of SIMs in realizable scenarios and produces multi-index model omnipredictors with eps^-2 prediction heads. Understand how this research advances the field toward proper omniprediction for general loss families and comparators, building upon previous work while dramatically improving sample complexity from eps^-10 to eps^-4.
