Microsoft has won one of the ten Patents for Humanity awards bestowed for the first time by the United States Patent and Trademark Office for providing machine learning tools that allow health researchers to better analyze large data sets.
There were five categories in USPTO's competition, a pilot program to promote humanitarian uses of patented technologies, which is intended to provide business incentives for patent holders to address global challenges in health and standards of living:
- Medicines & Vaccines
- Medical Diagnostics and Devices
- Food & Nutrition
- Clean Tech
- Info Tech
Microsoft entered Infer.NET, an inference framework devised by the Microsoft Research team in Cambridge UK into the final category and took one of the two awards. The other went to Sproxil for a solution that verifies that drugs and medicines supplied in developing regions of sub-Saharan Africa where there is a huge market in counterfeit drugs, are genuine.
Up against such competition you might wonder why a framework that allows rapid construction of complex Bayesian models and performs efficient inference within those models, won an award.
Two cases studies on Microsoft Research Connections provide the clue. One reports work done by John Winn, senior researcher in the Machine Learning and Perception group at Microsoft Research Cambridge and co-creator Infer.NET with the Wellcome Trust Sanger Institute on Understanding the Genetic Causes of Human Disease.
The second case study, Using Model-Based Machine Learning to Understand Childhood Asthma, a collaboration between Microsoft Research and the University of Manchester, aimed to build complex models that represent a broad range of important variables associated with asthma and to highlight the benefits of a model-based approach to the analysis of clinical data generally.
Microsoft makes Infer.NET freely available for non-commercial research purposes and helps outside groups use the tool effectively. It is currently at beta 2.5 and can be used with many .NET languages including C#, C++. Visual Basic and Iron Python to solve many different kinds of machine learning problems, from standard problems like classification or clustering through to customized solutions to domain-specific problems.
A new feature in Infer.NET 2.5 is Fun, a library turns the F# into a probabilistic modeling language for Bayesian machine learning. You can run your models with F# to compute synthetic data, and you can compile your models with the Infer.NET compiler for efficient inference.