Artificial Intelligence and Machine LearningYou may not know what machine learning is, but you most likely use it every single day. Do you watch Netflix? How about online shopping, are you an Amazon fan? Are you on Facebook? Do you have a bank account? If you answer yes to any of these questions, you’re benefiting from artificial intelligence and machine learning. For your reference, here are some everyday examples of machine learning in use:
● Mobile applications that can diagnose skin diseases
● Banks with algorithms that analyze spending patterns to detect fraud
● Facebook’s auto-tagging feature that knows who your friends and family are
● Email sorting that locates spam
● Netflix, Amazon, and other websites that have recommendations
● Virtual assistants like Siri, Alexa, and Cortana
Anyway, the goal of machine learning is for machines to learn without explicit programming; scientists want them to learn as we learn, for example, by using our past experiences to dictate our future actions. So instead of the step-by-step instructions that programming entails, this would be putting the ball entirely in the machine’s court; they would be given steps 1 and 2, and then the rest is up to them.
How Artificial Intelligence Allows Machines to ‘Learn’So, how exactly does machine learning work? Simple (well, not really simple, but simple to explain), it’s facilitated by the use of neural networks. For an idea of a neural network, think of how our brains work. They are, after all, made to mimic brain functionality. In essence, they’re like layers of neurons, and the more layers, the ‘deeper’ the network and the better they can learn. For example, let’s look at Digital Trends’ analogy comparing neural network learning to a factory line:
After the raw materials (the data set) are input, they are then passed down the conveyer belt, with each subsequent stop or layer extracting a different set of high-level features. If the network is intended to recognize an object, the first layer might analyze the brightness of its pixels. The next layer could then identify any edges in the image, based on lines of similar pixels. After this, another layer may recognize textures and shapes, and so on. By the time the fourth or fifth layer is reached, the deep learning net will have created complex feature detectors. It can figure out that certain image elements (such as a pair of eyes, a nose, and a mouth) are commonly found together.
From here, researchers specify what each output is, and correct any mistakes so that the machine can learn what to classify as X and not as Y.
The Four Ways Machines LearnNow that we have a general idea of how machines learn, let’s delve deeper and focus on the four ways machines learn.
Supervised learning is like learning from a teacher in the sense that the output is already known (students already know what they will learn). The work here comes in arriving at the output from a given input. For example, a machine learning from data sets may make a mistake in the process. The supervisor will then step in and correct it, allowing the machine to learn.
Unsupervised LearningUnsupervised learning is more like what people are aiming for when they think of computers ‘teaching’ themselves. As it’s unsupervised, there’s no training involved, no reference data, nothing. Comparing it to its supervised counterpart, imagine that the goal is for a machine to sort different shapes into groups (triangles, squares, pentagons, etc.). For supervised learning, you tell the machine this is a triangle, this is a square, and so on. Based on what you told it, it will then sort the rest of the shapes into their given categories. For unsupervised learning, the machine sees the same input, and based on what information it has available, will group them into their own categories. It won’t know what they are, but because they share a certain likeness, will know that they belong in the same category.
Semi-supervised LearningSemi-supervised learning lies between the two types of learning outlined above. There’s labeled and unlabeled data (more unlabeled data), and they work together to improve learning accuracy.
Reinforced learning uses rewards and connects actions with outcomes. Similar to how we learn — by trial and error — the machine will learn by directly interacting with its environment to achieve a certain goal.