Figure 4
From: Label-free imaging flow cytometry for analysis and sorting of enzymatically dissociated tissues
MLP screening. (A) Sketch shows general design of multilayer perceptrons. The input layer contains all pixels of the provided image. Each of the following \(k\) hidden layers contains \({n}_{i} (1\le i\le k)\) nodes. Each node represents a linear combination of the input values, which is modulated by an activation function (ReLU for the hidden layers and Softmax for the output layer). The output layer returns probabilities for each class of the classification task. (B) The scatterplot shows the inference time and number of trainable parameters of 396,521 different MLP architectures with \(k\)=1 (red), \(k\)=2 (orange), \(k\)=3 (blue), \(k\)=4 (magenta). Chosen models with identical inference time, but more trainable parameters compared to MLPNawaz are indicated by MLP1, MLP2, and MLP3.
