arXiv eess.SP· Saim Rehman, Muhammad Shafique·· 5 小时前AI 评分38
ORBITAL 提出面向嵌入式脑机接口的鲁棒性约束压缩框架
ORBITAL: Optimization of Robust Brain--Computer Inference Through Architecture-Aware Low-Precision Compression
AI 导读
ORBITAL 是一个鲁棒性约束的 EEG 解码器压缩框架，结合硬件可实现的结构化压缩、恢复感知工作点选择、多攻击评估与边缘端验证。在 BCI Competition IV-2a 的 9 名受试者、3 个随机种子上，S25-TRADES 在压缩 EEGNet 模型中取得最高平均 true-union 鲁棒性 23.06%，参数量减少 26.27%、MACs 减少 25.10%。
正文 · 原文
Abstract:Compact 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. Across nine subjects and three seeds on BCI Competition IV-2a, S25-TRADES achieves the highest observed mean true-union robustness among compressed EEGNet models (23.06% vs 21.27% for standard Full), with 26.27% fewer parameters and 25.10% fewer MACs. Matched controls identify TRADES as the main source of robustness recovery, improving S25 union accuracy by 6.74 percentage points over clean recovery (p=0.0039), while per-example analysis reveals boundary redistribution rather than uniform margin expansion. On Jetson Orin Nano, FP16 compression reduces board energy by up to 16.20% and latency by 8.00%. Despite 99.3% prediction agreement, native INT8 Q/DQ incurs about twice the FP16 latency and energy, demonstrating the need for joint robustness and hardware-aware optimization.
| Comments: | Submitted to IEEE ISCAS 2027 for Review |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Cryptography and Security (cs.CR); Machine Learning (cs.LG); Signal Processing (eess.SP) |
| Cite as: | arXiv:2610.36411 [cs.CV] |
| (or arXiv:2610.36411v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2610.36411 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Saim Rehman [view email]
[v1]
Mon, 26 Oct 2026 23:45:14 UTC (4,882 KB)
来源：arXiv eess.SP · arxiv.org