DIVINE : Coordinating Multimodal Disentangled Representations for Oro-Facial Neurological Disorder Assessment.
| Authors | Mohd Mujtaba Akhtar et al. |
| Year | 2026 |
| Venue | EACL 2026 |
| Paper | View on ACL Anthology |
| PDF | Download |
Abstract
Abstract not yet available in this stub. Read the full paper →
Engineering Breakdown
Plain English
This paper presents DIVINE, a system for assessing oro-facial neurological disorders (conditions affecting facial muscles and nerves) using multimodal data coordination. The approach disentangles representations across different input modalities—likely video, audio, or other sensor data—to improve diagnostic accuracy for neurological conditions like facial paralysis or dystonia.
Key Engineering Insight
The core technical contribution is coordinating disentangled representations across modalities rather than just fusing them—meaning the system learns separate, interpretable feature spaces for each modality and coordinates them meaningfully, which improves both accuracy and clinical interpretability compared to standard multimodal fusion approaches.
Why It Matters for Engineers
Medical AI systems need to be interpretable and auditable for clinical adoption. A disentangled representation approach lets clinicians understand what visual features (facial asymmetry, muscle tone) versus audio cues (speech clarity) the model is actually using for diagnosis, reducing the black-box problem that blocks healthcare deployment.
Research Context
Previous multimodal medical assessment systems typically used late-stage fusion or attention mechanisms that don't separate what each modality contributes. This work advances the field by bringing disentangled representation learning—proven effective in vision and NLP—into clinical multimodal analysis, enabling more robust and explainable neurological disorder screening.
:::tip Subscribe Get weekly breakdowns of papers like this in AI Letters - the newsletter for engineers building production AI systems. :::