The best definition for Machine Learning – in my opinion – is from the excellent book “Introduction to Machine Learning (MIT Press 2014, Ethem Alpaydin)” (Use it as a reference – this is not an easy “how to” book)
The goal of machine learning is to program computers to use example data or past experience to solve a given problem.
In general when we want to solve a problem on a computer, we need an algorithm to transform using a set of instruction into an output. Unfortunately for some problems we do not know how to program such algorithms – such as for e-mail spam detection or predicting customer behavior. In most cases we have the input and output available e.g. a set of e-mails for which some are marked as spam. Based on this data, we would like a computer (or machine) to automatically extract the algorithm necessary to perform the classification. The algorithm does not need to be perfect but needs to be a good and useful approximation.
The term machine learning is tightly coupled to the domain of analytics (or data science – see Data Scientist: the sexiest job of the 21st century ). Analytics is concerned with the discovery and extraction of useful business patterns or mathematical decision models from a specific data set. For this a number techniques can be used, depending on the practitioners background they will probably favor a technique from their respective domain:
- Regression, General Linear Models (GLMS), decision trees, etc … (originated out of the statistics domain)
- Machine learning algorithms such as support vector machines, neural networks, Bayesian methods, … (originated out of the computer science domain)
But why should you care about machine learning? I think the picture below shows you how the focus is shifting from traditional reporting (hindsight) to more advanced predictive and prescriptive analytics (foresight) which will provide business with more added value but also requires business intelligence specialist new competencies such as machine learning and data mining. Examples across industries vary but in general predictive analytics has the potential to change the way how businesses make decisions (I will take a look a more in depth definitely pick up Predictive Analytics – The power to predict who will click, buy, lie or die from Eric Siegel)
Microsoft Azure Machine Learning distinguishes itself from other platforms and tools by a number of different characteristics:
- Allows you to jointly build predictive models from anywhere in the world using only a web browser by making use of visual composition canvas (called Machine Learning Studio) using modules without requiring you to write code (although you can use R code snippets if you want). You can start quickly from existing sample experiments/models or you can share your own data experiments.
- Collaborative work together with anyone from anywhere using just your browser.
- Available as a cloud service, eliminating upfront costs fro hardware resources.
- The different modules allow you to author an end-to-end machine learning workflow starting with reading data, to training and validating your predictive model.
- Ability to deploy models as web services. You can quickly operationalize your models by converting them into web services and you even the ability to monetize your machine learning models using Azure Data Market.
- Microsoft Azure Machine Learning product page
- Machine Learning Blog
- Introducing Microsoft Azure Machine Learning (TechEd Europe 2014 recording)
- Microsoft Learning on Azure (AzureConf 2014 recording)
- Extensibility and R Support in the Azure Machine Learning Platform
- How to upload an R package to Azure Machine Learning
- Vowpal Wabbit Modules in AzureML
- Predict What’s Next: How to get started with Machine Learning Part 1
- Predict What’s Next: How to get started with Machine Learning Part 2
- AzureML : a short introduction