I understand the high level of the softmax algorithm but I’m a little confused how these decision boundaries are being generated; is it purely just for illustrative purposes to show when the activation values are past a certain point, the classification of some datum may change?
Those are plots of where each of the output classifier values are 0.5. Each one is a roughly pie-slice shaped region.
One way to generate this type of plot is to vary the x1 and x2 values over a grid, and draw colored contour lines between the locations where the value is 0.5 for each output unit.
1 Like