Research & Validation Program

Falsifiable Science for Multimodal Social Masking.

NDMS translates affective computing and neurodiversity research into a testable, reproducible system with physiological validation, ablation studies, and strict safety guardrails.

Explore Validation Domains EEG Knowledge Distillation

Methodological Rigor

Full Validation Program & Research Domains

All metrics and accuracy targets represent planned research goals unless supported by empirical dataset publication.

Construct Definition

Operationalizing social masking as deliberate or automatic suppression of spontaneous affect and effortful maintenance of normative facial/vocal cues.

Construct Rigor

Labeling Strategy

Multi-rater clinician annotations, self-reported retrospective effort tags, and task-induced cognitive/social stress protocols.

Ground Truth

Unimodal vs Multimodal Ablations

Systematic ablation tests comparing face-only, voice-only, unweighted fusion, and dynamic quality-gated fusion across datasets.

Ablation Science

Calibration & Discrimination

Evaluating expected calibration error (ECE), ROC-AUC, and precision-recall curves to ensure scores represent true probabilistic evidence.

Statistical Metrics

Reliability & Stability

Test-retest evaluation across repeated social interaction sessions to confirm temporal stability and low within-person variance.

Repeatability

Subgroup Fairness & Robustness

Audit slices across gender, age, ethnicity, regional accents, lighting levels, and diverse neurodivergent communication profiles.

Bias Controls

Latency & On-Device Execution

Profiling sliding window buffering, feature extraction latency (latency profiling / target to be preregistered), memory footprint, and mobile battery usage.

Edge Profiling

Human Factors & Utility

Qualitative user studies evaluating perceived agency, emotional clarity, fatigue recovery rate, and absence of anxiety/stigma.

User Benefit

Laboratory setting showing research equipment — Illustrative research photograph showing laboratory setup.

EEG Boundary & Knowledge Distillation

EEG is a Training Teacher—Not a Production Requirement.

A core scientific innovation of NDMS is using multi-channel EEG during initial laboratory training as a high-fidelity physiological "teacher."

Through teacher-student knowledge distillation, the student model learns to detect latent stress and masking signals using only audiovisual inputs—eliminating the burden of headwear or sensors for everyday end users.

Laboratory Teacher Model → Audiovisual Student Model

Evaluation Matrix

Planned Empirical Evaluation Framework

Validation domains and metrics will be explicitly preregistered in formal study protocols prior to clinical and public beta trials.

Calibration Error (ECE)Evaluated across confidence bins to ensure score calibrationTo be preregistered

Multimodal GainAblation comparison evaluating multimodal vs unimodal streamsTo be preregistered

Subgroup Fairness ParityDisparity audit across accents, demographics, and neurotypesTo be preregistered

Latency BudgetReal-time window processing on standard edge hardwareTo be preregistered

User Self-Regulation UtilityProspective evaluation of fatigue and masking self-understandingTo be preregistered