1

Topology-agnostic EEG foundation model that delivers state-of-the-art performance with 300× fewer FLOPs and 10× lower memory usage.

Bidirectional Mamba EEG foundation model that matches transformer accuracy while scaling linearly with sequence length.

Transfers EEG foundation models to ECG/PPG for BP estimation, reaching 1.57 mmHg MAE with 3.5× model compression via INT8 quantization.

On-device continual learning for wearable BMIs that reaches 92% accuracy with 1.6 min recalibration and 1 mJ training steps.

Lightweight imagined-speech decoder that runs on GAP9 at 17 mW using only eight EEG channels.

BERT-inspired EEG encoder that cuts seizure-detection false positives to 0.23 FP/h after subject-specific tuning.

Wearable ski-jumping coaching system with 92.7% phase accuracy running on GAP9 with multi-day autonomy.

In this paper, we introduce EpiDeNet, a lightweight seizure detection network, and a novel loss function (SSWCE) to address imbalanced datasets, achieving high accuracy, reduced false positives, and energy-efficient performance on low-power embedded platforms.

EEGformer, a compact transformer model for seizure detection on raw EEG traces, demonstrating efficient execution on MCUs with competitive accuracy.

In this paper, we present implementations of energy-efficient artifact detection algorithms on a parallel ultra-low power platform.