baseline4-DeepConvLSTMimp

DeepConvLSTMimp

*模型出处:Improving Deep Learning for HAR with shallow LSTMs 2021

模型结构

Model

数据集

  1. Opportunity

  2. Wetlab

  3. RealWorld (HAR)

  4. SBHAR

  5. HHAR.

对比实验

baseline:DeepConvLSTM

在多个数据集上对比了DeepConvLSTM和作者提出的单隐层的DeepConvLSTMimp网络,并且在单隐层的情况下,横向对比了隐藏单元的个数(128,256,512,1024)对结果的影响。

超参

  • batch_size: not mentioned
  • learning_rate: 1e-4
  • epochs: not mentioned
  • dropout: 0.5
  • batch_normalization: False

结果

Result

Highlights

  • 网络结构上一咪咪的改动只要足够多的对比工作也行啦~

我实现的DeepConv网络架构部分代码

*不放了…就在把DeepConvLSTM中间的lstm2删了


baseline4-DeepConvLSTMimp
http://example.com/2023/05/16/baseline4-DeepConvLSTMimp/
Author
John Doe
Posted on
May 16, 2023
Licensed under