MIDiff: Tackling Sparsity and Imbalance in Mobile Usage Generation via Multivariate-Imaging Diffusion

2026-07-15

AI Summary

Mobile usage data exhibits sparsity, heterogeneity, and inter-application imbalance, posing significant challenges for existing generative models to capture its underlying patterns effectively. To address this, this work proposes MIDiff, a multivariate imaging diffusion framework that first transforms sparse multivariate time series into correlation-aware images via the Cross-Gramian Angular Sum Field (C-GASF), then performs diffusion-based generation in the image space using a U-Net architecture enhanced with a triple attention mechanism. This design preserves temporal consistency and inter-variable dependencies during synthesis. The proposed method achieves state-of-the-art generation fidelity, with a discriminative accuracy (DA) as low as 0.1526—substantially outperforming the strongest baseline, ZITS-VAE (DA = 0.3476)—and produces synthetic sequences that are both more realistic and diverse.

Abstract

Mobile usage traces are critical for tasks such as user behavior prediction and app recommendation, yet their use is constrained by privacy restrictions and costly large-scale data collection. Although generative models perform well on general time series, their application to mobile usage data remains challenging because (i) limited user activity causes severe sparsity, (ii) heterogeneous variable types complicate joint modeling, and (iii) functional differences across apps create pronounced usage imbalance. To address these challenges, we propose Multivariate-Imaging Diffusion (MIDiff), a diffusion-based framework operating in an imaging space defined by Cross-Gramian Angular Sum Field (C-GASF). C-GASF transforms sparse multivariate sequences into correlation images, while MIDiff employs Triple Attention in a U-Net to preserve temporal consistency and variable dependencies. Experiments show that MIDiff achieves state-of-the-art performance across fidelity metrics. In particular, it obtains a Discriminative Accuracy (DA) of 0.1526, compared with 0.3476 for the strongest baseline, ZITS-VAE, demonstrating its effectiveness in generating realistic and diverse mobile usage traces. Our code is available at https://github.com/YilaiLiu-HKU/MIDiff.

Problem

• sparsity
• imbalance
• mobile usage generation
• heterogeneous variables
• time series generation

Innovation

• Multivariate-Imaging Diffusion
• Cross-Gramian Angular Sum Field
• Triple Attention
• Mobile Usage Generation
• Sparsity and Imbalance