Uncertainty Estimation in Time-Series Forecasting

​​Reliable uncertainty quantification in time-series forecasting is challenging due to temporal dependencies: observations are correlated over time, which violates the independence assumptions commonly used by many uncertainty estimation methods.

​This talk reviews modern algorithmic approaches developed to address this challenge and discusses practical strategies for constructing reliable prediction intervals for sequential data.

Speaker: Andro Sabashvili, PhD

​​Location: Terminal Abashidze street

Language: English