Machine Learning: What Is It?
The brain is an amazing creation. Who would’ve thought that such a small mass of tightly bound coil-like structures has in it embedded the whole consciousness of a human being. Till date, no one has ever fully understood how the brain functions, acting as a control centre covering every aspect of the human functionalities ranging from emotional to logical. While this may be true, a large number of attempts have been made at that, which though did not produce irrefutable theories, have at the least set us on the right track to understanding one of the most crucial parts of the human body. These attempts have given birth to a new field called Artificial Intelligence, or colloquially known as Machine Learning. Simply put ML is our endeavour to model a machine to think like the human brain and make decisions based on logic. While the thinking mechanism of the brain is a mystery, entangled with the right play of neural signals, various transformations of chemicals to electrical signals, the nervous system, and the body as a whole, the thinking of a machine has its roots deep in mathematics, which when progressively increased in complexity naturally corresponds to an improved performance of the machine.
But before we proceed, let us try to answer the question,”What is a machine?” This was one of the fundamental question put forth by Alan Turing, who proceeded to find the answer to this question in his famous paper, “Computing Machinery and Intelligence”, who finally put ‘digital computer’ on the mantle and went forward to answer the more immediate questions on the possibilities of existence of artificial intelligence.
To explain what Machine Learning is, allow me to borrow an explanation given by the Google team on this topic, in one of their introductory seminars to ML. The process fundamentally consists of three parts :
- Model – A model which describes a phenomenon, and gives an output accordingly.
- Parameters – The parameters consist of the components of the model, or those quantities which define the model, changing which alter the model proportionately.
- Learning System – This is the crucial part of the process, which gives the machine an ability to imitate the brain, or the part which is closest to the actual human learning phenomenon. Basically, this system compares the model output with the true output, and changes the parameters according to the error found, continually until the model predictions match the actual predictions with utmost accuracy, which is the essence of learning through experience.
It must noted that the model a machine follows is either initially fed in by human guidance or has to be deciphered by the machine itself. More will be seen on this below. The above-mentioned learning system forms the essence of what one may call Artificial Intelligence. Intelligence is a very abstract concept, which in itself is very hard to explain. It is a complex mixture of various set of structured processes, not fully understood by us, but related to each other, intertwined in a synchronous manner.
Now coming to manners in which this process can go about. There are generally three distinctions which cover this, but there also exist a whole lot of other techniques each with a small variation that fill this space. Limiting ourselves to generality, they are:
- Supervised Learning: When one thinks of supervision, one can imagine the need of a supervisor providing guidance that can serve as essential information to make sense of a process. The very same applies to a machine, where the learning is brought about by the a critical comparison of initially provided information with their respective outputs to generate a model, that requires the human touch for the first time in the form of training data, eventually leading to the machine being able to predict the output of another raw data set.
- Unsupervised Learning : Imagine yourself stranded in some faraway place, left to survive on your own by learning through bitter battles with not so good experiences, eventually leading to some form of order created by these learnings. While that might have been overplayed a bit, the machine is left in an identical situation, where there is no form of human guidance. The machine tries to see patterns in the raw data set, and make predictions after organising the data in an acceptable way. Now this kind of learning is used a frequently for the recognition of the patterns.
- Reinforcement Learning : As the name goes, this learning is achieved through reinforcement in the form of rewards. This learning optimises a machine to maximise a particular action through the notion of a cumulative reward, much like a promise of a treat or reward made to you for doing a particular action.
At this point, if you’re thinking, “But, how exactly does the machine learn?” it must be stated that there is no simple answer to the question. But a simplified illustration can lead you to the vicinity of the answer.
Consider a straight line initially made out of a given number of, say n, points. Now, in the case that additional information is provided to you pertaining to the inclusion of more points on the line, but the observed trend is that these additional points do not fit into the line perfectly. A possible solution to this problem would be the simple application of a linear regression to find the best-fit line that passes through most of the given points. In this case, we have essentially updated our initial model, which was a straight line, to a newer model, the best fit line that can accommodate all the given information. This is what the machine would do in this case, and this illustration can be extended to the application of more complicated mathematical, statistical, probabilistic techniques to a more difficult problems.
The possibility of the applications of Machine Learning are endless, with everything around us in this world serving as a potential automation experiment. After all, the world is a messy place, thus accentuating the possibility of pattern recognition through ML. But generally, we can find the use of ML algorithms in the following fields.
- Online Marketing and E-Commerce Platforms: These platforms can target customers and generate personalised advertisements based on the previous record of the customer through the use of ML algorithms.
- Banks and Financial platforms: Banks can predict defaulting patterns of customers, maximisation of investment opportunities through ML.
- Health care industries: Possible uses of machine learning are very important to these industries, with the scope ranging from the detection of malignant tumours based on previous data to the identification of a disease based on minute initial symptoms.
- Oil and Gas Industries: These industries can implement the ML algorithms to detect oil fields, improve the mining of oil through prediction of problems etc.
These are just some of the industries which frequently use Machine Learning in their day-to-day activities. But in fact, ML can be applied to a multitude of daily activities to streamline them and increase their efficiency. The field of ML, though drastically different from what it was a decade ago, still has a lot of ground to cover, and consequently there is still lot of hope left that one day, the mysteries of our brain will finally be unlocked.
Amateur Robotics is a wonderful field of science at the moment. Read more about it here.