It seems that we hear about artificial intelligence (AI) every day. What is it? AI refers to the capability of machines to perform tasks that typically require human intelligence. The term actually dates back to 1956. The extent of the use of the term AI has been heavily influenced by prevailing trends, causing it to go in and out of fashion in common language. The latest boom in the field of AI is driven by some extraordinary achievements of modern AI systems in performing tasks linked to image recognition, self-driving cars, and the ability of AI to beat the best human players at strategy games like chess, Go, and others. While AI systems can use a multitude of methods, the current developments have mostly relied on a specific technique called neural networks. The concept of neural networks in computers has existed since the inception of AI. However, with the increasing power of computers, larger neural networks, or deep neural networks, have become a reality. Let us discuss some of the key terms related to modern AI.
Artificial Intelligence (AI)
A machine is said to demonstrate artificial intelligence if the tasks it performs are considered to require human-like intelligence. On the other hand, machine learning and deep learning are specific methods through which machines/computers execute these tasks. Many modern AI systems use machine learning and deep learning, but not all of them do.
Machine Learning
Machine learning refers to a method where the computer is presented with a series of "questions" (or inputs) and their corresponding "correct answers" (or outputs). It then uses this information to predict the correct answer when presented with a new question. For example, an automated system for detecting credit card misuse can be trained on a variety of inputs like place and frequency of use, amount of purchase and others to automatically flag potential unauthorized use.
Deep Learning
Deep learning is a specific kind of machine learning that uses large neural networks and requires a large amount of data in comparison to typical machine learning to perform tasks that are complex and extensive. They consist of layers of interconnected nodes that process information. The more layers, the “deeper” the network, allowing them to perform more complex and sophisticated tasks. Deep learning uses a process known as training. The network then utilizes these trained connections to produce an answer when presented with a new problem.
A recent example of deep learning in ophthalmology was a computer showing many thousands of color retinal photographs. The computer was told only whether the photograph was of a man or a woman, and the subject’s age to one decade (i.e. 40-49 years, 50-59 years, etc.). After this deep learning, the computer was presented with a photo; the computer was able to tell whether the subject is a man or a woman and age to within one decade. What is remarkable about this is that no human, not even an experienced retina specialist, can do this. Another example of deep learning that we have probably all heard about outside the field of ophthalmology is the self-driving car, which collects and processes data on its surroundings from cameras and other sensors, interprets it, then decides what actions to take relative to the environment it is in.
ForeseeHome and SCANLY Home OCT
ForeseeHome monitoring uses a machine learning system. It has learned from past examples of patients testing on the device and their diagnosis of conversion from intermediate to wet AMD. When a new series of tests comes, it feeds the necessary parameters into this machine learning algorithm to determine if the patient has converted or not. Note that the ForeseeHome system does not use deep learning. The processes in SCANLY Home OCT uses deep learning. The neural network is trained using images that are marked by expert graders. The deep neural network connections are optimized using this training data. When a new image is presented to the system, it uses the trained network to determine if this data is present in a portion of the image or not.
What is the difference between artificial intelligence, deep learning and machine learning?
Description of terms we are likely to encounter related to AI. Note that there are versions of this Venn diagram on the internet that show machine learning and deep learning to be a complete subset of AI, which is not accurate.
Nishant Mohan, PhD, Vice President of Product, Notal Vision
