ORBITAL compresses EEG decoders under a robustness constraint
Original title: ORBITAL: Optimization of Robust Brain--Computer Inference Through Architecture-Aware Low-Precision Compression
S25-TRADES reached 23.06% mean true-union robustness versus 21.27% for uncompressed Full EEGNet.
ORBITAL combines structured compression, recovery-aware operating points, multi-attack evaluation, and native edge validation for compact EEG decoders. On BCI Competition IV-2a, across nine subjects and three seeds, S25-TRADES was the most robust compressed EEGNet variant, with 26.27% fewer parameters and 25.10% fewer MACs. Matched controls attributed most recovery to TRADES. On a Jetson Orin Nano, FP16 cut board energy by up to 16.20% and latency by 8.00%, while native INT8 used about twice the FP16 latency and energy.
- TRADES improved S25 union accuracy by 6.74 percentage points over clean recovery.
- That difference had p=0.0039.
- Per-example analysis showed boundary redistribution, not uniform margin expansion.
- Native INT8 still agreed with predictions 99.3% of the time.
From the original article
1 min read at arxiv.orgCompact electroencephalography (EEG) decoders enable embedded brain--computer interfaces (BCIs), but compression can alter adversarial robustness without proportional hardware savings. We present ORBITAL, a robustness-constrained framework combining hardware-realizable structured compression, recovery-aware operating-point selection, multi-attack evaluation, and native edge validation.
By Saim Rehman · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.