This document discusses methods for estimating the remaining useful life (RUL) of lithium-ion batteries using data-driven prognostics. It presents a bilinear kernel regression model that uses capacity fade data from batteries to predict RUL while accounting for noise in the training data. The model transforms the data using a kernel and employs LASSO regularization to provide sparse predictions and prevent overfitting. An experiment applies the model to capacity data from 8 test batteries and shows it can accurately estimate RUL even in the presence of noise in the training and test data.