A Path Towards Wearable Affective General Intelligence

Author: Solomon, Samuel Aaron

Year: 2025

Degree: Dissertation (Ph.D.)

Advisor: Gao, Wei

Committee Members: Emami, Azita; Anandkumar, Anima; Perona, Pietro; Gao, Wei

Option: Medical Engineering

Abstract

Artificial intelligence continues to support our daily decision-making tasks yet remains disconnected from our dynamic emotions driving these behaviors. Wearable technologies can supplement interactions with continuous emotion biofeedback, but existing models struggle to generalize across emerging biomarkers, platforms, and affective expressions. Here, we introduce a meta-analysis into embedding concurrent fragmented biosignals across 15 medical platforms, spanning five bodily locations, within a single profile that enables efficient and generalizable downstream affective analysis. We achieved this through a Lie manifold neural architecture that simultaneously reconstructs over 118,000 missing biometric details in 205 biomarkers and accurately forecasts 100 affective states across cohorts, questionnaires, and activities. We validated this framework across five datasets to propose a new skin-conformal, soft bioelectronic, affective computing platform that demonstrates closed-loop emotion modulation within thermal, audio, and visual interventions delivered through virtual, holographic, and conversational agents. Our framework establishes a new foundational bidirectional architecture for scalable, interpretable, and emotionally intelligent human-computer interactions.