Animikh Aich is a Deep Learning enthusiast, currently working as a Computer Vision Engineer. His work includes three International Conference publications and several projects based on Computer Vision and Machine Learning.
If you are a machine learning enthusiast and stay in touch with the latest developments, you would have definitely come across the news “Machine learning identifies links between the world's oceans”. Wait, we all know how complex it would be to analyse a concept such as oceans and their behaviour which would undoubtedly involve billions of data points associated with many critical parameters such as wind velocities, temperatures, earth’s rotation and many such. Doesn’t this piece of information gives you a glimpse of the wondrous possibilities of machine learning and its potential uses? And this is just a drop in the ocean!
As you move across this post, you would get a comprehensive idea of various aspects that you ought to know about machine learning.
Machine learning is a segment of artificial intelligence. It is designed to make computers learn by themselves and perform operations without human intervention, when they are exposed to new data. It means a computer or a system designed with machine learning will identify, analyse and change accordingly and give the expected output when it comes across a new pattern of data, without any need of humans.
The power behind machine learning’s self-identification and analysis of new patterns, lies in the complex and powerful ‘pattern recognition’ algorithms that guide them in where to look for what. Thus, the demand for machine learning programmers who have extensive knowledge on working with complex mathematical calculations and applying them to big data and AI is growing year after year.
Machine learning, though a buzz word only since recent times, has conceptually been in existence since World War II when Alan Turing’s Bombe, an enigma deciphering machine was introduced to the world. However, it's only in the past decade or so that there has been such great progress made in context to machine learning and its uses, driven mainly by our quest for making this world more futuristic with lesser human intervention and more precision. Pharma, education technology, industries, science and space, digital inventions, maps and navigation, robotics – you name the domain and you will have instances of machine learning innovations made in it.
Voice activated home appliances, self-driven cars and online marketing campaigns are some of the applications of machine learning that we experience and enjoy the benefit of in our day to day life. However, the development of such amazing inventions date back to decades. Many great mathematicians and futuristic thinkers were involved in the foundation and development of machine learning.
A glimpse of the timeline of machine learning reveals many hidden facts and the efforts of great mathematicians and scientists to whom we should attribute all the fruits that we are enjoying today.
This age laid the mathematical foundation for the development of machine learning. Bayes’ theorem and Markovs Chains took birth during this period.
Computers were recognised as machines that can ‘store data’. The famous Manchester Small-Scale Experimental Machine (nicknamed 'The Manchester Baby') belongs to this era.
Despite many researches and theoretical studies done prior to this year, it was the year 1950 that is always remembered as the foundation of the machine learning that we are witnessing today. Alan Turing, researcher, mathematician, computer genius and thinker, submitted a paper where he mentioned something called ‘imitation game’ and astonished the world by questioning “Can Machines Think?”. His research grabbed the attention of the BBC which took an exclusive interview with Alan.
The first artificial neural network was built by Marvin Minsky and Dean Edmonds this year. Today, we all know that artificial neural networks play a key role in the thinking process of computers and machines. This should be attributed to the invention made by these two scientists.
Though there were no specific terms till then for the things that machines did by thinking on their own, it was in 1974 that the term ‘machine learning’ was termed. Other words such as artificial intelligence, informatics and computational intelligence were also proposed the same year.
IBM developed its own computer called Deep Blue, that can think. This machine beat the world famous champion in chess, Garry Kasparov. It was then proved to the world that machines can really think like humans.
Back propagation is an important technique that machines use for image recognition. This technique was developed in this period of time.
Besides in 2014, a neural network developed by DeepMind, a British based company, developed a neural network that can access external memory and get things done.
In 2016, AlphaGo was designed by DeepMind researchers. It beat the world famous Go players Lee Sedol and Ke Jie and proved that machines have come a long way.
Scientists are talking about ‘singularity’ –a phenomenon that would occur if humans develop a humanoid robot that could think better than humans and will recreate itself. So far, we have been witnessing how AI is entering our personal lives too in the form of voice activated devices, smart systems and many more. The results of this singularity – we shall have to wait and watch!
To put it simply, machine learning involves learning by machines. It means computers learn and there are many concepts, methods, algorithms and processes involved in making this happen. Let us try to understand some of the more important machine learning terms.
Three concepts – artificial intelligence, machine learning and deep learning – are often thought to be synonymous. Though they belong to the same family, conceptually they are different.
It implies that machines can ‘learn on their own’ and give the output without any need of programming explicitly.
This term means machines can ‘think on their own’ just like humans and take decisions.
This involves creation of artificial neural networks which can think and act based on algorithms.
Quite simply, machines learn just like humans do. Humans learn from their training, experiences and through teachers. Sometimes they use knowledge that is fed into their brains, or sometimes take decisions by analysing the current situation using their past experiences.
Similarly, machines learn from the inputs given to them which tell them which is right and which is wrong. Then they are given data that they would have to analyse based on the training they have received so far. In some other cases, they do not have any idea of which is right or wrong, but just take the decision based on their own experiences. We will analyse the various concepts of learning and the methods involved.
The process of machine learning occurs in five steps as shown in the following diagram.
The steps are explained in simple words below:
The below examples will help you understand where machine learning is used in real time:
Voice based searching and call rerouting are the best examples for speech recognition using machine learning. The principle lies in translating spoken words into text and then segmenting them on the basis of their frequencies.
We all use this in day to day life in sorting our pictures on our Google drive or Photos. The main technique that is used here is classifying the pictures based on the intensity (in case of black and white pictures) and measurement of intensities of red, blue and green for coloured images.
Various diagnoses are increasingly made using machine learning these days. Here, various clinical parameters are input to the machine which makes a prognosis and then predicts the disease status and other health parameters of the person under study.
Machine learning helps in predicting chances of financial fraud, customer’s credit habits, spending patterns etc. The financial and banking sector is also doing market analysis using machine learning techniques.
Machine learning is all about machines learning through the inputs provided. This learning is carried out in the following ways:
As the name says, the machine learns under supervision. Let’s see how this is done:
Consider a shape sorting game that kids play. A bunch of different shapes of wooden pieces are given to kids, say of square shape, triangular shape, circular shape and star shape. Assume that all blocks of a similar shape are of a unique colour. First, you teach the kids which shape is what and then you ask them to do the sorting on their own.
Similarly, in machine learning, you teach the machine through labelled data. Then, the machine is given some unknown data, which it analyses based on the previous labelled data and gives the correct outcome.
In this case, if you observe, two techniques have been used.
As a further explanation,
Consider a kid playing with a mix of tomatoes and capsicums. They would sort them involuntarily based on their shape or color. This is an instantaneous reaction without any predefined set of attributes or training.
A machine working on unsupervised learning would produce the results based on a similar mechanism. For this purpose, it uses two algorithms as explained below:
The name itself says the pattern of this algorithm.
In reinforcement learning, there is no correct answer known to the system. The system learns from its own experience through a reinforcement agent. Since the answer is not known, the reinforcement agent decides what to do with the given task and for this it uses its experience from the current situation only.
Example: In a robotic game that involves earning the hidden treasure, the algorithm focuses on bringing out the best outcome through trial and error method. Mainly three components are observed in this type of learning: the user, the environment and the action the user is performing. The algorithm adjusts itself accordingly to guide the user towards the best result that can be achieved.
The diagram shown below summarizes the four types of learning we have learnt so far:
Machine learning is rich in algorithms that allow programmers to pick one that best suits the context. Some of the machine learning algorithms are:
To start the journey with machine learning, a learner should have knowledge of tools and libraries that are quintessential to designing machine learning code. Here is a list of such tools and libraries:
Machine learning is widely coded in Jupyter Notebook. It simplifies writing of Python code and embedding plots and charts. Google Colab is another free tool that you can choose for the same purpose.
The data gathering and preparation part of machine learning that we have seen in the stages involved in machine learning is taken care of by Pandas. This library:
NumPy supports all array based and linear algebraic functions needed while working on data, while SciPy offers many scientific calculations. NumPy is more widely used in many real time applications of machine learning as compared to SciPy.
This is a machine learning library that has an extensive collection of plots and charts. This library is a collection of many other packages. Of them, Seaborn is the most popular and is widely used to work on data structures.
These are known for their usage in Deep learning.
Besides algorithms, machine learning offers many tools and processes to pair best with big data. Various such processes and tools that are at hand for developers are:
Here is a list of five use cases that are based on machine learning:
As we have seen just now, machine learning is being adopted in many industries for the potential advantages it offers. Machine learning can be applied to any industry that deals with huge volumes of data, and which has many challenges to be answered. For instance, machine learning has been found to be extremely useful to organizations in the following domains which are making the best use of the technology:
Pharma industry spends billions of dollars on drug design and testing every year across the globe. Machine learning helps in cutting down such costs and to obtain results with accuracy just by entering the entire data of the drugs and their chemical compounds and comparing with various other parameters.
This industry has two major needs to be addressed: attracting investor attention and increasing investments, and staying alert and preventing financial frauds and cyber threats. Machine learning does these two major tasks with ease and accuracy.
By predicting the possible diseases that could affect a patient, based on the medical, genetic and lifestyle data, machine learning helps patients stay alert to probable health threats that they may encounter. Wearable smart devices are an example of the machine learning applications in health care.
Companies study the patterns that online shoppers are adopting through machine learning and use the results to display related ads, offers and discounts. Personalisation of internet shopping experience, merchandise supply panning and marketing campaigns are all based on the outcomes of machine learning results themselves.
Machine learning helps in predicting accurately the best location of availability of minerals, gas, oil and other such natural resources, which would otherwise need huge investments, manpower and time.
Many governments are taking the help of machine learning to study the interests and needs of their people. They are accordingly using the results in plans and schemes, both for the betterment of people and optimum usage of financial resources.
Machine learning greatly helps in studying stars, planets and finding out the secrets of other celestial bodies with far lesser investments and manpower. Scientists are also maximising the use of machine learning to discover various fascinating facts about the earth and its components.
Currently, machine learning is entering our lives with baby steps. By the next decade, radical changes can be expected in machine learning and the way it impacts our lives. Customers have already started trusting the power and comfort of machine learning, and would definitely welcome more such innovations in the near future.
Artificial Intelligence and Machine Learning have reached a critical tipping point and will increasingly augment and extend virtually every technology enabled service, thing, or application.
So, it would not be surprising if in the future, machine learning would:
Machine learning is quite different in its own way. While many experts are raising concerns over the ever increasing dependence and presence of machine learning in our everyday lives, on the positive side, machine learning can work wonders. And the world is already witnessing its magic – in health care, finance industry, automotive industry, image processing and voice recognition and many other fields.
While many of us worry that machines may take over the world, it is totally up to us, how we design effective, yet safe and controllable machines. There is no doubt that machine learning would change the way we do many things including education, business and health services making the world a safer and better place.