Machine Learning, Deep Learning , & Artificial Intelligence (A.I.) 101
It’s crazy to think that we’re currently in the year 2020. Online commerce is continuously growing, self-driving cars are becoming more common, people have easy access to vast amounts of information through portable supercomputers (smartphones), and we aren’t automatically labeled as crazies for speaking to our computers. Many of these achievements are thanks to the breakthroughs in tech with the help of the concepts and implementation of machine learning.
What is Machine Learning?
Machine learning is a set of methods that can detect patterns of data and then uses those patterns of data to predict future data or perform other kinds of decision making. More simply put machine learning gives computers the ability to learn without being explicitly programmed.
This kind of programming is done with input and output examples rather than just pure code, meaning that there needs to be large amounts of data to pull these inputs and outputs from. Machine learning falls within the scope of computer science but more specifically artificial intelligence.
What is Deep Learning?
Deep learning is a sub field of machine learning that uses algorithms inspired by the structure and function of the human brain called artificial neural networks. It was created by Andrew Ng, a global leader in AI, co-founder of Coursera and founding leader of Google Brain.
Deep learning uses a structure of interconnected layers: an input layer, intermediate layers (or hidden layers), and an output layer. The term deep refers to the intermediate layers, there can be as little as two layers to as many as one hundred layers and it’s where the learning takes place. Deep learning models are trained by using large sets of data and neural networks that learn directly from the data instead of manual input.
What is Artificial Intelligence?
Artificial intelligence is “the science and engineering of making intelligent machines”, as said by John McCarthy who coined the term back in 1955. AI doesn’t have one definitive definition but it can be summed up as a machine or computer program with the ability to solve problems the way a human would.
The original 7 aspects of AI:
1. Simulating higher functions of the human brain.
2. Programming a computer to use general language.
3. Arranging hypothetical neurons in a manner so that they can form concepts.
4. A way to determine and measure problem complexity.
5. Self-improvement.
6. Abstraction: defined as the quality of dealing with ideas rather than events.
7. Randomness and creativity.
Most artificial intelligence algorithms are expert systems. An expert system is a system that uses human knowledge and a computer to solve problems that normally require human expertise. The first non-expert system is AlphaGo. AlphaGo is the first computer program to defeat a professional human Go player and the first to defeat a Go world champion. It uses basic machine learning algorithms to play and win at the game Go, and those algorithms could be used as a basis to solve more complex or pressing problems such as climate modeling or disease analysis.
Weak AI v. Strong AI
A widely used, but uncommonly known artificial intelligence system is Apple’s Siri. Siri is a virtual assistant that uses speech recognition and natural language processing to convert human speech into text to classify it accordingly. And not to downplay any of Siri’s features that include a support for a wide range of user commands such as: performing phone actions, checking basic information, scheduling events and reminders, handling device settings, and searching the Internet, Siri is actually a form of weak AI.
There are two types of AI: weak AI and strong AI. A weak AI (also known as narrow or applied AI) focuses on a narrow scope of tasks, such as Siri. Siri is only useful if the user decides to interact with it. Weak AI is aimed at creating programs that can automate mundane tasks and its results usually don’t tell us anything about how humans think.
Strong AI focuses on creating programs that simulate how humans think and reason. The ultimate goal of creating a strong AI is creating a machine or program that can experience human consciousness instead of just simulating it. A strong AI hasn’t been achieved yet because humans are still trying to figure out what makes them human! If humans don’t yet completely understand how they function, how can they program computers to do the same?
How are machine learning, deep learning, and AI related?
Machine learning, deep learning, and artificial intelligence all fall within the realm of computer science. Machine learning algorithms are required in order to create artificially intelligent programs and deep learning is a sub field of machine learning. Deep learning uses artificial neural networks designed to mimic neurons in the human brain, which is how information gets passed throughout the brain. Instead of “neurons”, deep learning uses nodes that are arranged and interconnected in multiple layers. So artificial intelligence encompasses machine learning, and deep learning by default.
Different Types of Machine Learning
Machine learning is divided into three main categories: supervised, unsupervised, and reinforcement learning.
Supervised (or predictive) learning is the most widely used form of machine learning and it uses labeled data. What this means is that the machine already knows the results of the labeled training data and the machine simply learns how to get the correct results.
Labeled data requires people to identify and make judgements about unlabeled data. For example, a type of unlabeled data could be a picture of an animal, a cat or a dog. A person is then asked whether the photo contains a cat or a dog, and this is how labeled data is most often obtained.
A real-world application of supervised learning is the Google SmartASS (ad selection system), it’s a system that predicts the probability that you will click on an ad based on your search history. It acquires anything the user has ever typed into the search bar and then uses that data to produce ads that the user might be interested in clicking on.
Supervised learning can be broken down into two types: classification and regression.
· Classification is the organization of labeled data. There can be two classes (binary classification) or more (multiclass classification).
· Regression is the prediction of future outcomes based on the patterns of labeled data.
Unsupervised (or descriptive) learning only uses unlabeled data in order to find some form of structure or patterns within the data, this process is known as knowledge discovery. This differs from supervised learning because the machine isn’t programmed to look for specific kinds of patterns so there is no error metric to use. Unsupervised learning uses more difficult machine learning algorithms than supervised learning because it uses unlabeled data.
Unlabeled data is raw data that is obtained from the world around us such as pictures, videos, audio recordings, articles, news, tweets, etc. There is no label or “explanation” for each piece of data, it’s just the data itself. Since we know little to nothing about the data this also means that we won’t know what to expect of the output it will produce. This creates a less controlled environment since the machine is creating the outcomes for us.
It may sound a bit strange to have a machine classify and determine what exactly a piece of data is but it’s not that different than human and animal learning. Geoff Hinton, a famous machine learning professor at the University of Toronto, has said:
“When we’re learning to see, nobody’s telling us what the right answers are — we just look. Every so often, your mother says ‘that’s a dog’, but that’s very little information.”
A common form of unsupervised learning is clustering. Clustering is the analysis of patterns and groupings of unlabeled data. The machine sorts through the data to find some form of structure and begins to cluster the data into groups.
Reinforcement learning is a method that involves an agent and an environment. An agent is a computer program that has the capability to make simple decisions to solve problems, and they solve problems by interacting with other agents or with their environment. Reinforcement learning normally learns as it goes meaning the agent chooses between exploring or sticking to what it knows best, it’s a case of trial and error. This method allows the agents to maximize their performance based on the ideal behavior within a specific context. It’s useful for learning how to act or behave when given reward or punishment signals.
Wrap Up
Machine learning, deep learning, and artificial intelligence are all under the umbrella of computer science. To reiterate the Russian doll analogy, deep learning is a smallest doll nested in machine learning. Deep learning differs from machine learning because deep learning uses nodes that are connected through many layers (a faux human neural network) in order for the machine to test and learn from the data. Machine learning is the doll that holds deep learning and it’s a bit more broad since it encompasses many types such as: supervised, unsupervised, and reinforcement learning. They all allow for the computer to learn without being explicitly programmed. Finally, the biggest doll that holds both machine and deep learning is artificial intelligence. Artificial intelligence is the field of trying to create a computer or program think and solve problems like a human would. An AI with human consciousness still doesn’t exist yet, but with the help of machine and deep learning it doesn’t seem too far from our future.
