###### tags: `論文摘要` `ensemble` `LFP`

# Stacked Recurrent Neural Network for Decoding of Reaching Movement Using Local Field Potentials and Single-Unit Spikes
* 時間: 2017
* Conference: IEEE EMBS Conference on Neural Engineering
* Link: https://ieeexplore.ieee.org/iel7/8003578/8008262/08008440.pdf
* MLA: Fathi, Yaser, and Abbas Erfanian. "Stacked recurrent neural network for decoding of reaching movement using local field potentials and single-unit spikes." 2017 8th International IEEE/EMBS Conference on Neural Engineering (NER). IEEE, 2017.

## 概論
single unit activities (SUAs)具有良好的空間與時間特徵，故能提高BCI解碼表現，然而相較LFP不利長期解碼；故而如何兼取兩者特徵為該篇重點。**CTO**
僅有少篇文獻採取該作法。

## 方法
採用ensemble learning的概念將兩個不同decoder進行堆疊 (stack)，以進行權衡。
另有比較多種堆疊的方式。


LFP特徵方面選用LMP與5個頻帶(0–4, 7–20, 70–115, 130-200, 200–300)的特徵進行解碼。
LMP方面採256ms的moving average進行。
LFP時頻則於每個data window取256ms，且overlap為156ms的Hanning window進行。

firing rates (FRs)則以100ms為一個time bin，並取moving Gaussian window (256ms)且overlap為156ms用以平滑化FRs。