Entropy-Driven Curriculum for Multi-Task Training in Human Mobility Prediction

TL;DR

A unified training framework that integrates entropy-driven curriculum and multi-task learning to address mobility challenges and quantifies trajectory predictability based on Lempel-Ziv compression and organizes training from simple to complex for faster convergence and enhanced performance is presented.

Abstract

The increasing availability of big mobility data from ubiquitous portable devices enables human mobility prediction through deep learning approaches. However, the diverse complexity of human mobility data impedes model training, leading to inefficient gradient updates and potential underfitting. Meanwhile, exclusively predicting next locations neglects implicit determinants, including distances and directions, thereby yielding suboptimal prediction results. This paper presents a unified training framework that integrates entropy-driven curriculum and multi-task learning to address these challenges. The proposed entropy-driven curriculum learning strategy quantifies trajectory predictability based on Lempel-Ziv compression and organizes training from simple to complex for faster convergence and enhanced performance. The multi-task training simultaneously optimizes the primary location prediction alongside auxiliary estimation of movement distance and direction for learning realistic mobility patterns, and improve prediction accuracy through complementary supervision signals. Extensive experiments conducted in accordance with the HuMob Challenge demonstrate that our approach achieves state-of-the-art performance on GEO-BLEU (0.354) and DTW (26.15) metrics with up to 2.92-fold convergence speed compared to training without curriculum learning.

Figures (8)

• Figure 1: Three trajectories with heterogeneous complexity in the YJMob100K dataset Yabe2024, presented at the same spatial scale. The entropy reflects trajectory predictability, which will be explained in Section \ref{['sec:entropy1']} and \ref{['sec:entropy2']}.
• Figure 2: The entropy-driven curriculum learning pipeline. ① Real trajectories are augmented by mirroring and rotation. ② Augmented trajectories are ordered with increasing $H_{\textit{norm-LZ}}$ and subjected to progressively longer prediction horizon (Pho) to form a curriculum for pretraining. The pretrained model is subsequently finetuned using the raw dataset.
• Figure 3: Trajectory augmentation by mirroring and rotation.
• Figure 4: The multi-task prediction head. Losses are computed by cross-entropy.
• Figure 5: Feature interaction pipeline. The output shapes are denoted in gray.