Undergraduate Researcher (First Author)

Researching domain-generalizable meta-learning for physiological stress detection using EDA and HRV signals.

What I'm Doing

I am developing meta-learning architectures for stress detection that generalize across individuals without person-specific calibration. The model uses electrodermal activity (EDA) and heart rate variability (HRV) as inputs. I am learning a latent feature space that minimizes distribution shift between subjects.

Impact (Expected)

A domain-generalizable model would make wearable stress detection deployable without collecting calibration data for each user. This enables real-time closed-loop interventions for mental health and workplace wellness applications.

What I'm Learning

I am gaining experience with MAML and related meta-learning algorithms for few-shot adaptation. The physiological signals require preprocessing pipelines for artifact removal, resampling, and feature extraction (time-domain HRV metrics like RMSSD, frequency-domain power spectral density). I am learning about domain adaptation techniques and how to evaluate generalization across subjects using leave-one-subject-out cross-validation.

Key Highlights

• Researching domain-generalizable meta-learning architecture for physiological stress detection (EDA/HRV), utilizing latent feature optimization to minimize distribution shifts and triggering closed-loop interventions via signal-derived state policies.