Xylos brands

Predictive Analytics with Machine Learning

Machine Learning is a branch of data science that studies and creates algorithms that can learn from input data that we feed to the algorithm. We are interested in Machine Learning to construct a program or algorithm that can be used to make predictions: we train an algorithm with e.g. 100 input examples of both the question that we want to ask and the known answer, and then we can extrapolate the most likely prediction for new input data that the algorithm was not trained for. Interesting with Machine Learning is that the algorithm can (usually) be made more intelligent, with more accurate predictions, as it gets used more often.

Everyday usage examples of applications of Machine Learning are:

  • Spam classification - the spam filter has been trained with thousands of examples, the prediction we try to make on a new & incoming email is whether it is spam or not. By correcting the classification result in our mail client (marking a message as junk/not junk or spam/not spam), we improve the spam detection for the next mail that will come in.
  • Optical Character Recognition (OCR) - postal offices have been using Machine Learning for decades to automatically read destination addresses. The algorithm is trained with a few thousand variations of handwritten numbers, and afterwards is capable of detecting a new handwriting by extrapolating what it is learned for example on how the number "8" is typically written by humans.
  • Self-driving vehicles - Autonomous vehicles & aircraft typically involve a combination of different algorithms; the idea is that the car or aircraft "learns" from a human driver how to react in a curve or at a road crossing when driving around for a few hours, and then the algorithm can apply that knowledge on new roads and in new situations.

[caption id="attachment_22241" align="aligncenter" width="150"]Source: https://commons.wikimedia.org/wiki/File:Spam_can.png A can of Spam (Source: Wikimedia Commons)[/caption]

[caption id="attachment_22281" align="alignnone" width="150"]Source: https://commons.wikimedia.org/wiki/File:More_A%27s.jpg OCR (Source: Wikimedia Commons)[/caption]

[caption id="attachment_22261" align="aligncenter" width="150"]Source: http://theoatmeal.com/blog/google_self_driving_car Self-driving cars (Source: The Oatmeal)[/caption]

Machine Learning relies heavily on statistics and other mathematics to automatically tune the prediction algorithm (or "model") to the training data. Of course, one could also resort to manually implementing the prediction algorithm for the specific question in a traditional programming style. Even though this is perfectly possible for a single problem, the real power of Machine Learning is in its ability to adapt to new & incoming data; in that sense, Machine Learning can be used to automatically discover logic and patterns in data that would otherwise require tedious modelling and programming; the combination of multiple prediction algorithms at the same time can then be used to create what appears to be an "intelligent" or smart system.

[caption id="attachment_22282" align="aligncenter" width="300"]Source: http://www.wallpapermania.eu/wallpaper/say-hello-to-the-funny-robot An example of an "intelligent" system[/caption]

Public Cloud plays an important role in the growing adoption of Machine Learning, it greatly simplifies building your own algorithms and publishing them as a micro service API to use in your own applications. At Xylos Cloud Services, we work with Azure and Amazon Machine Learning, both which have their strengths:

Machine Learning: Amazon vs Azure

For a detailed discussion, you can have a look at our introductory cloud chat on machine learning, and a screen cast dedicated to Azure Machine Learning Studio.

Do you think the world will be taken over soon by intelligent machines? Or will we just stick to "smart" applications to make our lives a lot easier? Let us know what you think in the comments below!

Deel dit blogbericht

Also interesting for you

Laat een antwoord achter

Uw e-mailadres wordt niet gepubliceerd. Verplichte velden zijn gemarkeerd.

Breng jouw kennis en skills naar een hoger niveau

Schrijf nu in voor onze nieuwsbrief en krijg maandelijks:

  • Uitnodigingen voor Xylos' events & webinars
  • De laatste blogposts en cases
  • Nieuwste IT-trends