Welcome to Machine Learning History. After watching this video, you will be able to: Explain the motivation for machine learning. Describe significant discoveries that contributed to current machine learning technologies. List events that led to the rise of deep learning. Describe some machine learning methods discovered throughout history. Machine Learning technologies have been booming in the present world and have revolutionized industries in today’s world. The concept of machine learning is nothing new and many milestones throughout history have contributed to the wide applications of the technology that we see in the modern world today. We will be exploring the timeline of these milestones. In 1805, the mathematician Adrien-Marie Legendre published a paper on his discovery of the method of least squares, a method for finding the best line for a set of data points. Today, the least squares method is the standard approach for regression models. In 1950 Alan Turing proposed "The Turing Test”, a method to determine if a machine can think like a human. The test essentially requires a machine to have a conversation with a human being by text, and if after 5 minutes the human is convinced that they are talking to another human, then the machine has passed the test. Arthur Samuel, an engineer from IBM, formulated the term "machine learning" and, in 1952, developed a computer program for playing checkers. This was a milestone because this appeared to be the world's first self-learning program. The ideas embodied in this program were the beginning of a widely-used field of machine learning known as reinforcement learning. In 1957, a psychologist and professor at Cornell University, Frank Rosenblatt, submitted a paper proposing the "Perceptron", a system for pattern and shape recognition that is similar to how neurons work in the biological brain. The perceptron is a single-layered neural network that is used to make predictions. If a prediction the system made was wrong, then the system would adjust itself to make a better-informed decision next time. His ideas are widely seen as the foundations of deep learning. In 1967, the nearest neighbor algorithm was conceived. The nearest neighbor algorithm was used for classification, and the way the algorithm worked was, it assigned an unclassified point to the same class of the closest set of previously classified points. The nearest neighbor algorithm was a starting point of algorithms that don't make assumptions about the sample data. In 1970, a Finnish master's student, Seppo Linnainmaa, published a paper on the reverse mode of automatic differentiation, a way of efficiently computing how fast a function is changing using a graph. And he implemented this method in computer code. Many believe this discovery was the first step in creating an algorithm widely used for training a neural network called backpropagation. In 1979, Japanese computer scientist Kunihiko Fukushima published his work on the neocognitron, a system that can recognize visual patterns through learning. The neocognitron is inspired by two types of cells discovered in the part of the brain that processes visual information called simple cells and complex cells that were discovered decades earlier. The neocognitron consists of "S-cells" that work similarly to simple cells by extracting features, and "C-cells" that work similar to complex cells and are responsible for recognizing patterns despite their deformations or changes in positions. His work inspired what we know as convolutional neural networks today. In the 1990s, the work in machine learning shifted to a data-driven approach. Scientists and researchers created programs that analyzed large amounts of data to find patterns and learn from the results. The growth of the internet also created an ever-growing availability of digital data. This leads to the legendary event in 1997 when IBM's Deep Blue beats the world champion in chess. In 2006, Geoffrey Hinton, a cognitive psychologist and computer scientist, published a paper on how one can use “deep” neural networks to obtain state-of-the-art results. This was the first time the word “deep” was used in a machine learning context. Geoffrey Hinton, then, developed the term "Deep learning.” Once deep learning's potential and ability was showcased, technology giants started investing heavily in deep learning, and there was a rise in deep learning startups. In 2011, IBM's Watson computer beats two of the most successful human contestants in a game of Jeopardy. Jeopardy requires contestants to answer questions on various topics based on the clues received. This victory marked a monumental moment in the history of machine learning because of IBM Watson’s ability to understand large amounts of human knowledge and marked another step in making machines think like a human. A field of machine learning called natural language processing helped drive the Jeopardy win. IBM research, then, developed Project Debator, the first AI system that could successfully debate with humans on complex topics. In 2019, the 2016 World debating championships grand finalist and 2012 European Debate champion Harish Natarajan challenged Project Debator in a debate that ended in a draw. For a machine to be able to debate with humans, it must be able to identify relevant arguments and their quality. This was a milestone in making a machine that can master human language and make decisions like a human. In 2020, OpenAI released a groundbreaking natural language processing algorithm that could generate human-like text with a given prompt. This algorithm is called cGPT-3 and is considered one of the most advanced natural language models in the world. The quality of the text produced is so high that it makes it very hard to determine whether or not the text was actually written by a human. Machine learning technologies that we know of today are built upon almost a century of various advancements in technology. This field is continuously being explored and advanced due to its vast applications and possibilities in areas such as healthcare, finance, and marketing. Today, it is responsible for some of the most notable advances in technology such as self-driving cars. In this video, you learned about: Milestones throughout history in the field of machine learning. Important machine learning methods and their discoveries. The potential usage of machine learning in the present world.
