“The neural network is this kind of technology that is not an algorithm, it is a network that has weights on it, and you can adjust the weights so that it learns. You teach it through trials.”
-Howard Rheingold, American critic and writer
At a recent American Board of AI in Medicine (ABAIM) course, an attendee asked a simple yet complicated question:
“Is deep learning considered a supervised or unsupervised learning?”
Deep learning involves the use of complex models that exceed the capabilities of machine learning tools such as logistic regression and support vector machines, but these deep learning models are essentially “function approximators”.
Deep learning uses supervised learning in situations such as image classification or object detection, as the network is used to predict a label or a number (the input and the output are both known). As the labels of the images are known, the network is used to reduce the error rate, so it is “supervised”.
Neural networks, on the other hand, can also be used to cluster images based on similarities. One can extract the features with a neural network, then deploy an unsupervised methodology such as k-means clustering. A neural network can be in the form of a semi-supervised deep neural network.
In addition, autoencoders are neural nets that can be used for image compression and reconstruction via a latent space representation of compressed data; in short, it outputs whatever is inputted. These autoencoders are considered self-supervised learning neural nets.
Finally, reinforcement learning with neural networks can be used, and was the methodology behind DeepMind and its victory in the game Go.
Therefore, deep learning can be supervised, unsupervised, semi-supervised, self-supervised, or reinforcement, and it depends mostly on how the neural network is used.