Machine Learning #1 — Introduction
Hi, my name is Vedant Mathur, and in this blog I’ll be walking you through my pursuits and explorations in the revolutionary field of Artificial Intelligence (AI), Machine Learning and Big Data.
In a world as enveloped and infatuated with technology as ours, there are many technological innovations that capture the minds and hearts of the populations, but one field stands out amongst them all. One field, whose ambition is second only to it’s potential, is growing at an unfathomable rate, and is set to mark the dawn of the new age. That field, is Artificial Intelligence.
As a field, Artificial Intelligence investigates and delves deep into the very foundations of our psychologies and philosophies, working to formulate what classifies as intelligence, and what does not. The enigma, of how to determine when a computer can actually ‘think’, is a question that plagues even the greatest of computer scientists and philosophers, and as we approach a clear-cut answer to that question, we can finally unveil a new wave of technology — Smart Computers.
While a lot of software we encounter these days masquerade as ‘featuring’ AI, at its purest sense, Artificial Intelligence looks to mimic and mirror our cognitive functions.
Now as a field, AI is huge. Everything, from Data Science, to Machine Learning, to Natural Language Processing (NLP), all fall under the overarching umbrella that is Artificial Intelligence. And so in that sense, AI isn’t really a field. It’s the seed that has sprouted into the other fields.
So, as I explored the endless realms of Artificial Intelligence, one field within it really stood out to me was Machine Learning. The more I read about it, the more it hooked me. Just the very thought of training a computer to ‘learn’, and almost think independently, truly lit up my mind. I thought of the endless reach a field such as this would have; the implications of Machine Learning pour into medicine, defense, finance and so much more. Therefore, I realized when studying AI, a field that truly encompasses and takes to heart the notion of independent ‘thinking’, was indeed Machine Learning.
So as I had my heart set on Machine Learning, I then began to explore the fundamental questions to which it is founded upon. It dawned upon me that to truly achieve any success in Machine Learning, we must first conquer the underlying question — How does one define ‘learning’? And as we dig deeper into this question, it becomes clearer and clearer that the field of AI has as much to do with philosophy as it does with computing; to try to qualitatively lay out what it means to ‘learn’, demands a serious exploration of our own psychologies. Fortunately, Tom M. Mitchell, a renowned American computer scientist, seems to have come out on top, offering a formal definition for what Machine Learning encompasses —
A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E.
Now, at the surface this may rub off as a cryptic and incoherent jumble of words; after all, he has thrown at us some terminology, like ‘experience E’, or ‘performance measure P’, which now may seem foreign to us, but in time these concepts will gradually set into your mind and cement into foundational ideas of Machine Learning.
To better illustrate what exactly Michell was trying to communicate in his informative, yet cryptic statement, I’ll introduce an example, where all these foreign terms will be better illuminated upon.
Problem Example — Classifying Tumors as Benign or Malignant
Let’s say we’re given data that compares the size of a tumor to whether or not it is benign (non-fatal), or malignant (potentially fatal). So, for instance, we may have a tumor that’s 2 mm, and that corresponds to it being benign.
Can we, as computer scientists, employ the practice of Machine Learning to actually learn from this data, and be able to predict whether a given tumor is either benign, or malignant.
Now, let’s take a few steps back and touch up with Tom M. Mitchell’s definition of computer learning. He referred to a task T, which in this instance would be predicting the state of the tumor (i.e. Benign, or Malignant). He then referenced the experience E, which in this case simply alludes to the data we feed to the algorithm. Finally, the performance measure is simply a means to evalutate the success and accuracy of the algorithm.
So, in essence, for the computer to be ‘learning’, its performance measure should rise as we feed in more data. If we tackle this notion in the frame of this example, the accuracy of the predictions should be getting sharper and more accurate as we feed in more data of tumor sizes and states.
Should this hold true, then congratulations, you have a working piece of Machine Learning software!
From this example, I hope you have all been able to better grasp what the essence of Machine Learning is. In addition to how it offers an explanation for Mitchell’s formal definition, it also subtly showcases the indisputable reach of this field. In this instance, we saw how Machine Learning can be employed to actually help classify tumors as benign or malignant, which in the real world is incredibly important, and outlines how this field can actually save lives.
So that’s all for this post. In my next one, I’ll be delving into the two ‘big’ branches of Machine Learning — Supervised Learning, and Unsupervised Learning.
Thanks for reading!