When someone says Machine Learning, it simply oozes cool and has a definitive sci-fi feel to it. However, you might be surprised to learn that you already know the fundamentals of machine learning! Machine Learning as a subject is fundamentally based on the shoulders of two other subjects—Mathematics and Computer Science. If you are someone who knows probability, statistics and has a firm grasp on linear algebra, you can be fairly confident that you’ll be able to grasp the concepts of machine learning. Now, let’s take a look at the inner working of machine learning.
Working of Machine Learning
The fundamental concept of machine learning is built on the fact that much of what we consider to be intelligence is based on probability, more so than logic and reasoning. As bizarre and counter-intuitive as this might sound, take a moment to think about the various situations in your life and try to correlate them. When you want to go from point A to point B, you probably think of the fastest route between them. When you play a board game, you try to see which move will help you win the game. Consider any such situation and you will see that probability plays a very big role in the decision-making process of humans.
Now, when it comes to computers, we know that they’re very good at making calculations. This was realized by scientists as far back as the 1950s. They understood that, with enough data, digital computers would be great at making estimations of probability. Unfortunately for the pioneering researchers of AI, the time in which such a revolutionary idea arose was not a time when it could be fully explored. Computers had not yet become powerful enough to run such novel ideas and be put to the test. Even so, their founding principles were right on the dot and these principles form the foundation upon which modern AI has been built.
Deep Neural Network
Companies like Google, Facebook, and Amazon use Machine Learning on all the data they can acquire from their customers. This is done to optimize user experience and preferences. One particular method of Machine Learning that all companies use is Deep Neural Networks. The concept of deep neural networks is based on an idea designed by Warren McCullough, Walter Pitts and Frank Rosenblatt related to neural networks in the 1950s. Although today’s neural networks are much more complex than the primitive and formative ones of yesteryear, the main idea is still the same and the idea is as follows. The best way to estimate a given probability is to break the problem into discrete, bite-sized pieces of information, which were coined neurons by McCullough and Pitts. The hunch that McCullough and Pitts had was that if a group of such neurons was connected in a way similar to that of a human brain, then different models could be built to learn different things.
To understand a neural network, let’s consider an image with a face in it. Now, if we were to have a primary deep neural network, this neural network would have several thousand nodes. Each of these thousands of nodes would be stacked up in layers. The first thing each node in the first layer of a neural network would look for is a line or a curve. Once a preliminary analysis is done, the second layer of the neural network would look for even more advanced shapes, like a circle. In the third layer, multiple parameters would be searched for, such as a dark circle in a white circle, which is how the human eye would be recognized in a deep neural network. When the algorithm finally reaches the neurons of the final layer, each neuron is able to identify advanced shapes. Based on the results of the last neurons, the algorithm can estimate whether it is a face or not.
Applications of Machine Learning
Ever since computers were invented, linguists and computer scientists have tried to make them recognize speech and text. This method by which a computer can realize written language or speech in a coherent and logical way is known as NLP or Natural Language Processing. However, over the past several decades, machine learning has largely surpassed rule-based systems, thanks to everything from support vector machines to hidden Markov models to, most recently, deep learning. Apple’s Siri, Amazon’s Alexa, and Google’s Duplex all rely heavily on deep learning to recognize speech or text and represent the cutting-edge of this important field.
The next area where Machine Learning is used very widely is Image Processing. When Rosenblatt first implemented his neural network in 1958, he first tested it using the images of dogs and cats. Ever since then, AI researchers have been obsessed with the topic. By necessity, much of that time was spent devising algorithms that could detect pre-specified shapes in an image, such as edges and polyhedrons, using the limited processing power of early computers. Thanks to modern hardware, however, the field of computer vision is now dominated by deep learning. When a Tesla drives safely in autopilot mode, or when Google’s new augmented-reality microscope detects cancer in real time, it’s because of a deep learning algorithm.
One of the final fields where Machine Learning is paramount (and also a personal favorite) is Robotics. What makes our own intelligence so powerful is not just that we can understand the world, but that we can interact with it. The same will be true for robots. Computers that can learn to recognize sights and sounds are one thing; those that can learn to identify an object and determine how to manipulate it is another thing altogether. However, as one can imagine, speech recognition is a difficult challenge, and touch and motor control are far more difficult skills to master. For all their processing power, computers are still remarkably poor at something as simple as picking up a shirt. The reason why picking up an object like a shirt isn’t a piece of cake for robots is because it involves several simultaneous tasks. First, you need to recognize a shirt as a shirt. You then need to estimate how heavy it is, how its mass is distributed, and how much friction is present on its surface. Based on those guesses, you will need to estimate where to grasp the shirt and how much force to apply at each point of your grip, a task made all the more challenging because the shirt’s shape and distribution of mass will change as soon as you lift it. A human does this trivially and easily, without a second thought, but for a computer, uncertainty in any of those calculations compounds across all of them, making it an exceedingly difficult task.
Although the applications of Machine Learning vary greatly, these examples should give some insight as to how deeply and profoundly it has already impacted industries and our lives.