SoftmaxWhat is the Softmax function and why is it needed?
|
|
The purpose of Softmax function is to adjust the outputs of a convolutional neural network (CNN) so that they sum to 1. This means the model output values are scaled so they can be thought of as relative "probabilities" that a certain category was predicted by the network.
The figure on the right demonstrates the use of Softmax in the final stage of a brain tumor classification CNN illustrated in the previous Q&A.
Here, the raw output data [−0.1, 3.8, 1.1, −0.3] has been transformed into relative "probabilities" by Softmax.
The figure on the right demonstrates the use of Softmax in the final stage of a brain tumor classification CNN illustrated in the previous Q&A.
Here, the raw output data [−0.1, 3.8, 1.1, −0.3] has been transformed into relative "probabilities" by Softmax.
In brief, the Softmax functions uses exponentials (ezi) to convert individual components of the output vector into positive numbers. For a given vector element (zi), Softmax transforms this value into a probability by first raising it to an exponential, then dividing by the summed exponentials of all the elements in the original output vector. As an example, the Softmax value for the second vector element z2 = 3.8 is calculated as
and so forth for the other elements in the vector. The Softmax function is highly nonlinear in that it tends to push the higher raw data values toward 1 and the smaller values toward 0.
References
Keisan Online Softmax Calculator. Available at this website.
Softmax Function. Wikipedia. The Free Encyclopedia.
Keisan Online Softmax Calculator. Available at this website.
Softmax Function. Wikipedia. The Free Encyclopedia.
Related Questions
What is a convolutional neural network (CNN) and how does it work?
What is a convolutional neural network (CNN) and how does it work?