Active Research August 2025 - Present

Real-Time Neuromorphic Biosignal System for Adaptive Learning

A closed-loop neuroadaptive system that adjusts content delivery based on real-time cognitive load inference using neuromorphic hardware

Overview

This project builds a complete human-in-the-loop adaptive system that monitors learner cognitive state in real-time and dynamically adjusts audio playback speed to optimize learning outcomes. The system demonstrates co-adaptation between human and machine.

Closed-Loop

Continuous sensing → inference → adaptation cycle

Human-in-the-Loop

Both learner and system adapt over time

Neuromorphic

Ultra-low-power SNN inference on Dynapse chip

System Architecture

Closed-Loop Neuroadaptive System Learner Cognitive State Sensors EEG PPG | EDA 256 Hz | 64 Hz LSL Sync Feature Extraction Power | HRV | SCR DYNAP-SE Neuromorphic SNN Inference μW power Load Class Audio Speed Closed-Loop Feedback

Figure 1: Complete system architecture showing multimodal sensing (EEG, PPG, EDA), Lab Streaming Layer (LSL) synchronization, SNN inference on Dynapse I, and adaptive content control.

My Role

System Architect & Lead Developer
  • Concept & Design: Designed the complete human-in-the-loop co-adaptation paradigm
  • Full-Stack Implementation: Built streaming acquisition, real-time feature extraction, and adaptation logic
  • Hardware Integration: Configured and deployed SNN on Dynap-SE neuromorphic processor
  • User Studies: Designed experiments to evaluate system efficacy and user experience

Real-Time Processing Pipeline

LSL Streaming Architecture Time → EEG 256Hz PPG 64Hz EDA 4Hz LSL Sync Timestamp Align Adaptation Logic Cognitive Load Class Policy Slow Down Keep Pace Speed Up Thresholds: High Load Optimal Low Load Personalized thresholds adapt over sessions

Figure 2: (Left) Lab Streaming Layer synchronization of EEG, PPG, and EDA with latency-aware buffering. (Right) Modality-specific update rates and feature-age auditing for robust decision making.

What's Innovative

This work introduces a novel human-in-the-loop co-adaptation paradigm where both the learner and the system adapt together over time:

<100ms Sense-to-Adapt Latency
3 Modalities Fused
Co-Adapt Human + System

Unlike traditional adaptive systems that only modify content, our approach enables bi-directional learning where system policies improve with user feedback while users adjust to system behavior—enabling personalized, long-term adaptation.

Live Dashboard

Real-Time Monitoring Dashboard Live Biosignals EEG PPG EDA Features Connected Buffer: 2.3s Cognitive Load Classification 72% Medium History: Adaptation Control Audio Speed 0.85x Adjustments (last 5 min) ↓ 3 slow ↑ 1 fast

Figure 3: Live monitoring dashboard showing real-time biosignals, extracted features, cognitive load classification, and adaptation history.

Related Projects