Bridging the Gap Between Training and Inference for Spatio-Temporal Forecasting

TL;DR

This paper introduces a curriculum learning strategy called Temporal Progressive Growing Sampling to improve spatio-temporal forecasting by reducing training-inference discrepancies, leading to better long-term dependency modeling.

Contribution

It proposes a novel training approach that gradually transitions from supervised to less-supervised learning, effectively bridging the gap between training and inference in sequence forecasting.

Findings

Outperforms baseline models on two datasets

Better captures long-term dependencies

Reduces error accumulation during inference

Abstract

Spatio-temporal sequence forecasting is one of the fundamental tasks in spatio-temporal data mining. It facilitates many real world applications such as precipitation nowcasting, citywide crowd flow prediction and air pollution forecasting. Recently, a few Seq2Seq based approaches have been proposed, but one of the drawbacks of Seq2Seq models is that, small errors can accumulate quickly along the generated sequence at the inference stage due to the different distributions of training and inference phase. That is because Seq2Seq models minimise single step errors only during training, however the entire sequence has to be generated during the inference phase which generates a discrepancy between training and inference. In this work, we propose a novel curriculum learning based strategy named Temporal Progressive Growing Sampling to effectively bridge the gap between training and inference for spatio-temporal sequence forecasting, by transforming the training process from a fully-supervised manner which utilises all available previous ground-truth values to a less-supervised manner which replaces some of the ground-truth context with generated predictions. To do that we sample the target sequence from midway outputs from intermediate models trained with bigger timescales through a carefully designed decaying strategy. Experimental results demonstrate that our proposed method better models long term dependencies and outperforms baseline approaches on two competitive datasets.