Neural Networks and TensorFlow - Deep Learning Series [Part 3]

In this third lesson on deep learning with TensorFlow we're briefly going to discuss softmax output.

We use softmax output in projects when we have to work with output that is made of multiple mutually exclusive values. In this case the values are the labels.

Ok, you may not understand what I'm saying. So, let me try to be more specific.

In a simple deep learning project where we train a model to recognize an animal, the output that we're going to look for is binary: animal or non-animal. Even more specifically, when we train a model to recognize cats, it will see images that are labeled with 'cat' or 'not a cat'. This is a binary output.

When we deal with a softmax output, we have more than two values and these values are mutually exclusive. We could have for example more than 2 animals that we wish to recognize: cat, mouse, horse, eagle, etc. And the image we'd be providing the classifier would contain only one animal. Or we could use more complex algorithms in which we'd have more animals in one image.

This is probably the essence of softmax output. Please watch the video for my complete explanation. In future lessons we'll be implementing softmax output in TensorFlow.

To stay in touch with me, follow @cristi

---

Cristi Vlad Self-Experimenter and Author