Dr. Isabelle Guyon, of New York University, is a highly-cited informatics researcher who currently facilitates competitions in machine learning to apply research to data modeling. You may think of predictive models as tools to help weather forecasting or stock market predictions. But amazingly enough, the same type of mathematics is also applicable to information retrieval with an Internet search engine, voice and face recognition, and tools to aid medical diagnosis. For example, in her recent research Dr. Guyon directed her interest to gesture recognition from video data, which is applicable to interpreting sign language used in communication between hearing impaired people and for signalling from a distance, in silent environments, noisy environments, or under water. Last year, Dr. Guyon collaborated with CERN physicists to bring to the attention of the machine learning community the problem of predicting which events in a collider indicate the occurrence of certain particles, such as the famous Higgs Boson. Through her nonprofit group ChaLearn (Challenges in Machine Learning) Dr. Guyon directs research into these important applications by designing “challenges” to provide incentive for scientists to think about these compelling problems.

This interesting and unique turn in Dr. Guyon’s career came with the realization that collaboration between an open pool of contributors would efficiently breed enriched innovation. After contributing multiple key inventions to the machine learning field herself (she is a co-inventor of a famous algorithm called “Support Vector Machines”), Dr. Guyon was struck by the thought that if she presented problems that she was trying to solve to a large number of scientists in the form of a competition, someone might improve upon her method, leveraging her initial contribution to push for the most cutting-edge technology. Since this epiphany, Dr. Guyon has opened the door for scientists of any age or professional level -- from bright undergraduates to casual hobbyists -- to explore and generate novel machine learning systems crucial to our current age of information. For years these competitions, garnering up to 300 teams at a time, have directed the attention of the scientific community to unique problems they might not have realized needed solutions, and have provided an organic method of teaching future computer scientists the tools they need to approach and tackle such problems on their own for many more years to come.

A few current projects include:

• Looking at People: Through human pose recovery, action/interaction recognition, and gesture recognition, Dr. Guyon is indulging the creation of machine learning practices that will revolutionize the job interview process. By using a motion-sensing camera, such as Kinect, recruiters can analyze and select applicants with an added layer of effectiveness, reducing wrong hiring decisions and unemployment.

• Discovering Neural Network Structures: Recovering the exact circuitry of the brain (up to 100 billion neurons) is a daunting tasks. However, with new technology of opto-genetics and fluorescence imaging, neuroanatomists can now read and write the brain and recovering the neural structures can be facilitated with machine learning. A data model that could explain the brain’s wiring would mean possible advances in Alzheimer’s and epilepsy research, as it would possibly point out blocks in neuron flow and other anomalies.

• Automatic Predictive Models: The Automatic Machine Learning (AutoML) challenge series aspires to create a machine capable of learning from examples without any human intervention. The methods of the winners of each round will be made publicly available and will save enormous amounts of human effort in data science. This challenge series received initial support from Microsoft. ChaLearn intends to pursue it by introducing problems of increasing difficulty to push research in the field.