Time Series Prediction under Distribution Shift using Differentiable Forgetting
This addresses adaptive modeling for time series prediction in non-stationary environments, but appears incremental as it builds on existing weighted empirical risk minimization approaches.
The paper tackles time series prediction under distribution shift by framing it as a weighted empirical risk minimization problem with a forgetting mechanism, and proposes a gradient-based learning method for the forgetting parameters, which speeds up optimization and allows for more expressive mechanisms.
Time series prediction is often complicated by distribution shift which demands adaptive models to accommodate time-varying distributions. We frame time series prediction under distribution shift as a weighted empirical risk minimisation problem. The weighting of previous observations in the empirical risk is determined by a forgetting mechanism which controls the trade-off between the relevancy and effective sample size that is used for the estimation of the predictive model. In contrast to previous work, we propose a gradient-based learning method for the parameters of the forgetting mechanism. This speeds up optimisation and therefore allows more expressive forgetting mechanisms.