Machine Learning

Machine learning: A basic explanation. 

Many companies tend to use machine learning and artificial intelligence interchangeably. And usually, that is not a problem. But sometimes, it’s not.  

Quoting Andrew Moore, Dean of the School of computer science, Carnegie Mellon University: “Artificial intelligence is the science and engineering of making computers behave in ways that, until recently, we thought required human intelligence.”  

So, to say, AI is the over-arching feel whereas, machine learning is one of the ways this artificial intelligence is developed in which the computer learns from its past experience.   

Again, from Tom M. Mitchell, computer scientist, professor at Carnegie Mellon: “Machine learning is the study of computer algorithms that allow computer programs to automatically improve through experience.  

Example: YouTube taking data like what videos you watch and for how long to suggest what you might want to watch next.   

Machine learning mainly adopts 3 systems:   

1. Unsupervised learning: In this one, the computer does not need the help of a coder and it uses a technique called “clustering” in which it takes all the given inputs and tries to find patterns in the inputs and then divides them into categories based on those patterns. An example would be the YouTube suggestions feature mentioned above.  

2. supervised learning: In this obviously a coder is present, who feeds labeled data to the computer and the computer then tries to find a function or equation of that data. The coder then again feeds labeled data and the computer matches its previous equation and does necessary adjustments. This continues until the equation is accurate enough to understand whether any unlabeled data matches the equation. An example of this would be the filters found in messenger or Instagram. The coders train the system by continuously feeding the computer different images of the human face until the computer can understand by itself which is a human face and which is not  

3. reinforcement learning: This method can be simply put as trial and error. The computer continuously tries various paths to reach the outcome. Paths favorable to the outcome are rewarded and the paths which are not, are penalized. Obviously, whether the paths are favorable or not is determined by the coder. Eventually, it understands which actions to perform for which type of tasks. An example of this one would be a self-driving car. The goal or outcome is to keep the car inside the lane. The computer tries to keep the car in the lane by trying various options at its disposal like steering, braking, and accelerating. Every time it goes out of lane it has to start over. Eventually, it will learn what it has to do to stay on lane.  

So, until now we just learned how the machine learns. Now we will talk a bit about how these processes happen in our devices. All the above learning techniques have different sets of algorithms that help the computer achieve the desired outcome. One of the more familiar algorithms is the “Neural Network”. An example would be: suppose a picture of a cat. The neural network would take the picture through various filters to tweak it slightly so that the computer has an idea of what a cat looks like. Then the computer will be able to distinguish whether the original photo was that of a cat or a dog. Regardless of the algorithm the programmer chooses to use, it requires a lot of mathematical computation, which most of the time the CPU or GPU in our devices are capable of doing, but it is not the most efficient way. This is where “AI processors” as they are called, enter the chat. Things like Tensor and scalar cores and vector extensions are just specialized processors that are built just for these types of tasks and are much more efficient at it. If you ever see any presentation of a new processor or CPU you will hear these names used quite a lot in the presentation.  

To end off, let me give a few real-life examples of machine learning you can see right now on your phone.  

1)The list of suggested apps in your app library is based on data regarding the apps you used on previous days around that time of the day. Suppose you always watch YouTube at night. So, your phone will suggest YouTube when you open the app library at night.  

2)When you click a picture, the phone automatically adjusts the white balance, color correction, sharpening, and HDR based on similar photos clicked by professionals.  

3) Real-time translation, Augmented reality, image and speech recognition, biometric recognition, and the list goes on and on.  

Bottom line, machine learning helps machines go through tons and tons of data that are available nowadays. This can help an entire industry to get work done quickly or help you get through your daily challenges a bit more easily.