Machine Learning: Get ready to be Dumbstruck
When we ‘Google’, how is it that it presents us with the most relevant result? How come Facebook always seems to know the kind of news feed we will be interested in? How can an online training course automatically pause when the person viewing it looks away? How does your Gmail identify spam mails among all the mails that you are receiving? If you are wondering about these questions, then the answer you are looking for is machine learning or predictive analysis. Millions and millions of gigabytes of data are being created on real-time basis and these companies are using machine learning to make sense of them.
Machine learning technologies use algorithms to explore data, analyze, study and identify patterns with which a realistic prediction of the outcome can be made. Now, this might have sounded simple but let me assure you that it is not. With the amount of data being generated every minute, the task of figuring out anything sensible from it is not humanly possible, or not manually possible. This is why companies like Google are spending tons and tons of dollars into advancing the technology that is doing this for them. Google, LinkedIn, Facebook, Pandora are all companies that became successful mainly because of their good command of the machine learning technology. Why shouldn’t they be, after all, identifying the patterns and utilizing them literally makes their services better. Is it not a win/win situation for all? The companies are happy, the customers are happy; thanks to this thing that makes the ‘Big Data’ more understandable.
Types of Algorithms
Algorithms are not created keeping the motto one size fits all, and they cannot be. This means that in machine learning, various types of algorithms, based on the type of input available, are being used. Some of them are:
- Supervised Learning
These algorithms are trained with inputs whose desired outputs are known (labelled). So, they generalize the process and use this to generate an optimum output to previously unseen inputs.
- Unsupervised Learning
Here, a desired output is not known (unlabelled) and so the algorithms analyze the given input or data so as not to generalize it but to identify a pattern and thus generate an output.
- Semi-supervised Learning
As the name suggests, these algorithms are trained to handle both labelled and unlabelled inputs.
- Reinforcement Learning
These are intelligent algorithms that execute actions to change the environment it functions in so as to understand and learn how the environment responds to the actions.
Big Data was being generated for decades now, but it was not until now we were able to do something meaningful with them. As mentioned before, to analyze the big data was not a task that was manually possible (of course in smaller amount it can be done, but not at the ‘big data’ magnitude). But now, this is changing. The success of Google search engine will be the perfect example of this change. Google has changed and evolved a lot over the years and a major chunk of the credit goes to the advancement in data mining and machine learning. Google’s voice search and software that does face recognition are all in one way or other related to machine learning. Facebook’s new algorithm, based on reports, takes into account about 100,000 factors to determine what shows up in your News Feed. But what Google or Facebook or LinkedIn does with the technology can only be considered as the tip of the iceberg when it comes to the possibilities that machine learning has opened. There are tailor made ads, news feeds, offers and more to begin with, but what are some of the other possibilities?
It was recently revealed that Featurespace, a UK based technology firm, is using behavioral data gathered from gaming sites to analyze, reason, and figure out ways to combat the increasing online gambling addiction in the country. Now, this is one use of machine learning that not many would have thought about before. The company is also using its machine learning algorithms to analyze customer buying patterns to spot anomalies or fraudulent behavior.
Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory and Center for Wireless Networks and Mobile Computing are working on a computer system (Remy) that has the ability to automatically generate TCP (Transmission Control Protocol) congestion-control algorithms. TCP regulates the rate at which computers send data, thus preventing network congestion. According to the researchers, Remy is a machine-learning system that, unlike typical manmade TCP congestion-control algorithm which has a handful of rules, can have more than 150 distinct rules, thus making the internet faster and better.
Now, there are many more uses of the technology but these were a few that caught my attention recently, and seemed worth mentioning. Keep an eye on the field and I can guaranty that you will be dumbstruck by the developments that are happening real-time.
Big Data is not new, but the tools that make us capable of analyzing it better and faster are. In the future with the exponential growth of IOTs (recently Gartner’s predicted that IOT will soon have 50 billion machine connecting around 2 billion users) and customized healthcare, the importance and prominence of machine learning will be epic. The possibilities are endless and the scope is vast for the technology. Some experts are already predicting that machine learning will be one of the biggest fields of interest in another 3 – 5 years. So, better tighten your seatbelts and ready yourself for the ride to a better future.