Re: Shape and Time Distortion Loss for Training Deep Time Series Forecasting Models

Manjot Singh, Yiyu Wang

Abstract: In this paper, some results from DILATE(Distortion Loss including shape and time) are reproduced and additional studies on hyperparameter explorations, loss function and different neural network models are done as a part of NeurIPS Reproducibility Challenge. The comparison between DILATE, Euclidean Loss (MSE) and Huber Loss has been shown when training Deep Neural Networks on non-stationary time series forecasting problems.

Track: Ablation

NeurIPS Paper Id: https://openreview.net/forum?id=r1ld_NBxIB