Simplifying How Machines Learn

Machine learning is like any other sense which develops immensely with time and keep enhancing its predictability with experience.

Let’s consider our childhood, our teachers used to tell us about numbers like even-odd, prime- composite etc. etc.

We used to learn the logic and based on that we used to predict. Similarly, I told myself that those numbers which can be completely divided only by 1 or itself are prime, those which can be divided by others as well are composite.

One day I was asked a question about ‘1’ for the very first time if its prime? I really got confused, I said its prime but I knew am wrong so asked my teacher next day,  then I came to know that this is indeed a special one.

‘1’ is neither a prime nor a composite. So, I updated my understanding that any number which has two divisors 1 or self are prime and ones with more than two divisors are composite.

Machine Learning is absolutely the same.

This goes out to everything we see around and how we classify and predict them to be in certain category.

Above mentioned scenario in Machine Learning is called Classification and it is a supervised learning.

Logic driven approached - Based on logic (definition), we are trying to classify a number whether it’s prime or composite.

My teachers were my supervisors and they used to correct my understanding (like definition of prime and composite in above case).

Machine learning model is our knowledge base which require regular upgrade by the data my teacher used to provide.

This knowledge base (data base or set) is called training data in Machine learning.

Without continuous inputs from my teachers, used to predict new objects and occasionally made mistakes especially when it did not fit into my existing definitions.

This data is called testing data.

Model Accuracy- Based on training data, I classified 6 out 10 scenarios correctly, and then my model accuracy would be 60%.

Machine Learning Algorithms- There were 40 students in our class. They gathered and learned same definitions but some of them double clicked it with their cousins or tuition teachers and hence they developed their own understanding, and predicted 8 out 10 correctly. Then some students acted as a different machine learning algos and their model accuracy for this data-set was 80% or 100%.

The hidden pattern identification and mapping from the data-set is what help Machines learn to automatically predict effortlessly, just like we do.

Machine Learning is just data science, only thing that is required is how simply we can relate the ultimate output of the exercise, how precisely we can go into details of it  and how effectively we can build a self learning mechanism with correct absorption. 

Disclaimer: Its coming out of my study on Neural Networks way back in college days. Intended is to create some interest and simplify.