By Tejas Venugopal
How are they related to each other? How do we teach machines to perform tasks?
The world of AI (Artificial Intelligence) is vast. Really, it’s huge. One can spend many years of their life learning and prodding and poking and still just end up scratching the surface. And all the jargons surrounding the field do not help either. Where does one even start trying to understand this beast?
There are many articles on the internet trying to explain AI Artificial Intelligence or on machine learning mastery using mathematics and theory. For someone just trying to make sense of this world, this approach is just wrong. The best way to explain any topic is through analogies to the real world. It’s much more relatable that way.
So, read on for a simple everyman’s explanation of machine learning.
I will also use ‘Tech Bubbles’ in a few places to give a better technical explanation of the concepts.
AI artificial intelligence is just an umbrella term, a vague idea which describes a class of problems and approaches to solve those problems.
This seems like such a random combination of people working together. However, this is what it takes to solve the big problems of today.
Machine learning is a part of AI which deals with learning to optimize algorithms using data.
So, let’s start with the heavily overused word, machine learning. To understand machine learning, one needs to understand human learning. How do humans learn? and specifically how do babies learn? Understanding this is the key to understanding machine learning.
Machine learning is just an umbrella word to lump in a lot of different concepts. When a machine ‘learns’ something, we call it machine learning. A machine here usually means a software program. For e.g. A machine can learn to recognize objects in the real world (from photographs). Here’s a ball, Here’s a toy, Here’s a dog, Here’s the floor. A baby can do the same, usually much faster. Both of these are examples of ‘learning’
Machine learning or traditional machine learning refers to using algorithms to model/fit the data. The final machine learning models become a function approximation from the input to the output. F(x) = y, where x is the input e.g. an image of a ball and y is the output e.g. ‘ball’. Here F is some function which transforms x into y. Machine learning algorithms such as logistic regression, naïve Bayes, Random forests, Support vector machines, etc. are popular for traditional machine learning
This is another word that’s really popular. There’s nothing to fear. Deep learning is a type of machine learning. You see how it’s an umbrella word now?
Deep learning is machine learning where we use more sophisticated algorithms like neural networks instead. Deep learning is a way of solving the same problems, but with approaches which are inspired by the human brain. But ultimately, it’s trying to achieve the same goal. ‘Learn’. It’s used as a powerful tool to solve some of the hardest AI problems. E.g. Natural language processing, Machine translation, Question answering, Object recognition, Language modeling, etc.
Deep learning uses neural networks which have more than 1 hidden layer of neurons. Another difference between traditional machine learning is that the features are selected by the network during training. Certain features of the input may be more useful than others, and neural networks use the concept of backpropagation of the loss to tune the weights of these features and every layer in between.
But how does the machine actually learn? And how good a teacher/parent do we need to be to teach it. Can we be the attentive parent and spend all our time on it? Can we afford to be a bit distracted? Will the machine still learn? Or can we be completely lazy?
Stay tuned for more in our next blog
You can have a look at these: