Primary sourcearXiv cs.LG
View sourceEvent ·Multi-source
Diff-Logic:用于边缘EEG低延迟分类的差分逻辑门网络
该研究提出Diff-Logic,将神经网络编译为纯布尔电路,通过位运算在CPU上执行。在四个EEG数据集(包括二分类痴呆检测和三分类情绪识别)上,与MLP和BNN基线对比,复杂度从50k到500k参数。痴呆检测中Diff-Logic达到80.2% Macro F1,比MLP高6.8%。在7W Nvidia Jetson Orin Nano单核CPU上,情绪识别任务中MLP性能略优但延迟是Diff-Logic的2.3倍,模型大小是14倍。Diff-Logic在模型规模扩大10倍时推理时间几乎不变,最大复杂度下速度提升达2.9倍。
Current read
这论文把神经网路编译成纯布尔电路,在边缘设备上跑EEG分类,延迟低还省电。痴呆检测准确率比MLP高6.8%,值得搞BCI的人看看。
3 sources47° AI heatLast updated 2026/7/20 16:49:30
Evidence
3 sourcesConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.