Table 6 The network architectures used for the feature extraction modules \(\phi _1\), \(\phi _2\). k denotes kernel size, f the number of filters in the convolutional layers and M the final embedding dimension. \(\mathbb {I}\) denotes the unit interval [0, 1].

From: Unobtrusive detection of Parkinson’s disease from multi-modal and in-the-wild sensor data using deep learning techniques

Feature extractor

\(\phi _1(\mathbf {x})\)

\(\phi _2(\mathbf {x})\)

Input

\(\mathbf {x}_{1k} \in \mathbb {R}^{3\times 500}\)

\(\mathbf {x}_{2k} \in \mathbb {I}^{500}\)

3-DOF acceleration

HT, FT hists

Layer 1

Conv1D \(k=8, f=32\)

Dense \(500 \rightarrow 100\)

LReLU (\(\alpha = 0.2\))

LReLU (\(\alpha = 0.2\))

MaxPool \(k=2\)

Dropout \(p=0.1\)

Layer 2

Conv1D \(k=8, f=32\)

Dense \(100 \rightarrow 50\)

LReLU (\(\alpha = 0.2\))

LReLU (\(\alpha =0.2\))

MaxPool \(k=2\)

Dropout \(p=0.1\)

Layer 3

Conv1D \(k=16, f=16\)

Dense \(50 \rightarrow M\)

LReLU (\(\alpha = 0.2\))

MaxPool \(k=2\)

Layer 4

Conv1D \(k=16, f=16\)

 

LReLU (\(\alpha = 0.2\))

 

MaxPool \(k=2\)

 

Layer 5

Flatten

 

Dense \(320 \rightarrow M\)

 

Output

\(\mathbf {h}_{1k} \in \mathcal {R}^M\)

\(\mathbf {h}_{2k} \in \mathcal {R}^M\)

Back to article page