Can we teach computers to learn?

Computers are machines that can do calculations faster than humans. But can they be taught to self-learn and identify patterns? This question is answered by machine learning. Machine Learning identifies patterns through experience. It uses past data to come up with patterns which can be used to predict the future.

ML is an interdisciplinary field, it borrows from philosophy, statistics, information theory, and biology to name a few. So, we see a lot of jargon that is common to these fields. worry not, we’ll learn them as and when required. But the underlying thought remains the same. Just the way a kid learns from his/her surroundings we make a system learn from past data (this is called training). So, in machine learning, learning can be defined as an improvement of a system’s performance with experience. We’ll refine this definition as and when we progress.

What is ML?

Machine Learning is a collection of algorithms which unearth patterns in data. The patterns can be as simple as normal rules, such as, if a person buys eggs and butter he is likely to buy butter, to very complex equations which includes trigonometric ratios and differential equations. The aim of a machine learning algorithm is to improve the results of doing certain tasks given experience. For example, consider the driverless cars experiment that Google’s company Waymo, Uber and other car companies are doing. How is ML used in these scenarios? Let me explain, A human drives the car and a complex software program learn the human’s driving patterns. The patterns can be slow down at an intersection, stop at red light, blinkers while changing lanes etc. Later the driverless car uses the patterns learned and navigates the road.

Okay, I get it, but where can I use ML?

Where can ML be used?

Machine Learning can be used in places where past experience can be used to learn a task. It goes like this, imagine that you are teaching a machine how to make coffee. Your target is to make the best coffee for your customers. How do you decide the best coffee? Every day you let your machine make coffee and serve it 10 customers and ask them how the coffee is. If 8 out of 10 people say I like the coffee then the training for coffee making task is complete and you can use the coffee making machine to serve coffee to a larger audience.

In summary, ML can be used for tasks where past experience can be used to make future predictions.

Why ML?

ML has a lot of cool applications. It is used to teach cars how to drive (Waymo) — driverless cars, teach machines how to play games such as chess and checkers, teach drones how to fly and a lot of other really cool applications. In ML, we come up with a mathematical equation from the past data. We tell the algorithm broad rules that it can use to come up with the equation such as the equation should be linear. This equation can be used to predict the future and also to interpret the past. We’ll delve deeper in coming episodes.

Good, but where can ML be not used?

Coming up in the next issue, Limitations of ML and ML algorithm classification.