Here is the definition from investopedia:” deep learning is an AI function that mimics the workings of the human brain in processing data for use in detecting objects, recognizing speech, translating languages, and making decisions. Deep learning AI is also able to learn without human supervision, drawing from data that is both unstructured and unlabeled”.
However, when it comes to health there is an inherent problem with unlabeled data. It allows unsupervised learning with identified patterns. The problem with health and its solutions, that the doctor needs to be able to explain his decisions; this is part of the black box, a major problem in the application of algorithms in health.
In AI in Dermatology all the different subsets can be used and the most common learning mechanism is through Machine learning which uses Convolutional Neural Networks, a type of artificial neural network
Comment
As a dermatologist we can interpretate deep learning as a purely visual experience of looking at a lesion an extracting more complex and abstract features in order to make a diagnosis and elaborate a management plan, taking into account the circumstances for each patient.
The first phase is a purely automatic electrical stimulation of cells into the retina, the elaboration of a signal and a passage through the optic nerve. Some fibers will go to autonomous parts of the brain such as the cerebellum, hindbrain and pineal gland.
However, the interesting part is the relay of these electric signals through the optic radiations to the primary visual cortex. For now, the localization of the pure electrical signal is precise and mapped in the cortex.
From now the relay is made to the associative visual cortex such as V4 and V5 which mediate color and movement for example as well as more complex interactions with the other loci of the brain. All these are interactions between interneurons with a modulation of synapses, and this constitutes part of the human learning process….AI through wired connections is different but the concept of parallel-distribution processing can be understood with this analogy.