This research adapts foundation model techniques to tabular data for detecting empathy from visual cues, outperforming traditional methods.

• Addresses privacy concerns by working with extracted features rather than raw video
• Demonstrates 8.2% performance improvement over classical machine learning approaches
• Introduces a novel architecture combining transformer blocks with MLP layers
• Validates results on human-robot interaction datasets

For healthcare applications, this advancement enables more accurate empathy assessment in therapeutic settings, improves patient-doctor interaction analysis, and enhances empathetic capabilities in medical care robots and assistive technologies.

Tabular foundation model to detect empathy from visual cues