A new paper introduces HiPAD, a unified fusion network for depression detection that is submitted to every sub-track of the MPDD-AVG 2026 challenge. The challenge spans two age groups—Elder and Young—and three modality configurations: gait alone, audio-visual alone, and the two combined. HiPAD is designed as a single model that can handle all six of these settings, rather than requiring separate systems for each condition.

The abstract describes the network's scope but does not provide technical details. The full title mentions a hierarchical personality ablation and the calibration limits of domain-adapted LLMs, suggesting that the work also explores how personality traits and large language model calibration affect detection performance. However, these aspects are not covered in the available abstract.

Because only one source is available, there is no independent comparison to note. The paper is posted on arXiv under identifier 2610.14961, and the abstract indicates that the work is a direct entry into the MPDD-AVG 2026 benchmark.