Joint optimization of linear and nonlinear models for sequential regression

We investigate nonlinear regression and introduce a novel approach based on the joint optimization of linear models. In order to capture both characteristics in sequential data, we model underlying data as combination models, where optimize models jointly minimize final error. As model, employ differentiable version boosted decision trees. use well-known SARIMAX model. Our is generic so that any or can be readily employed provided they are differentiable. By this optimization, alleviate underfitting overfitting problems modeling data. Through our experiments synthetic real-life demonstrate significant improvements over individual components well combination/mixture methods literature.