Table 8.

The identified best models for all benchmark datasets using resPsoCnn-PB-GB. TB indicates a transitional block which contains a single 1 × 1 convolutional layer and RB indicates a ResNet block which contains two convolutions, as indicated in Figure 3. FC indicates a fully connected layer.

Dataset	Structure
MNIST [3,52]	TB($c_{in}=1$ $c_{out}=177$) + RB(177 × 4 × 4) + RB(177 × 4 × 4) + RB(177 × 6 × 6) + AveragePool + TB($c_{in}=177$ $c_{out}=175$) + RB(175 × 6 × 6) + RB(175 × 6 × 6) + RB(175 × 5 × 5) + RB(175 × 3 × 3) + AveragePool + FC
MNIST-RD [53,54]	TB($c_{in}=1$ $c_{out}=161$) + RB(161 × 5 × 5) + RB(161 × 7 × 7) + RB(161 × 6 × 6) + RB(161 × 6 × 6) + RB(161 × 4 × 4) + RB(161 × 5 × 5) + RB(161 × 7 × 7) + TB($c_{in}=161$ $c_{out}=115$) + RB(115 × 5 × 5) + RB(115 × 7 × 7) + RB(115 × 5 × 5) + RB(115 × 6 × 6) + RB(115 × 3 × 3) + RB(115 × 7 × 7) + RB(115 × 4 × 4) + AveragePool + FC
MNIST-RB  [53,54]	TB($c_{in}=1$ $c_{out}=153$) + RB(153 × 4 × 4) + RB(153 × 6 × 6) + + RB(153 × 4 × 4) + RB(153 × 3 × 3) AveragePool + TB($c_{in}=153$ $c_{out}=183$) + RB(183 × 4 × 4) + RB(183 × 6 × 6) + RB(183 × 7 × 7) + AveragePool + FC
MNIST-BI [53,54]	TB($c_{in}=1$ $c_{out}=136$) + RB(136 × 4 × 4) + RB(136 × 3 × 3) + RB(136 × 5 × 5) + RB(136 × 3 × 3) + AveragePool + TB($c_{in}=136$ $c_{out}=136$) + RB(136 × 6 × 6) + RB(136 × 5 × 5) + RB(136 × 5 × 5) + RB(136 × 3 × 3) + RB(136 × 3 × 3) + RB(136 × 3 × 3) + AveragePool + FC
MNIST-RD+BI [53,54]	TB($c_{in}=1$ $c_{out}=150$) + RB(231 × 5 × 5) + RB(231 × 5 × 5) + RB(231 × 7 × 7) + RB(231 × 3 × 3) + AveragePool + TB($c_{in}=150$ $c_{out}=98$) + RB(120 × 4 × 4) + RB(120 × 6 × 6) + RB(120 × 6 × 6) + RB(120 × 5 × 5) + AveragePool + FC
RECTANGLES-I [53,54]	TB($c_{in}=1$ $c_{out}=195$) + RB(195 × 3 × 3) + RB(195 × 6 × 6) + RB(195 × 3 × 3) + AveragePool + TB($c_{in}=195$ $c_{out}=85$) + RB(85 × 7 × 7) + RB(85 × 5 × 5) + RB(85 × 3 × 3) + AveragePool + FC