Watson kicks off cognitive computing research

The recent avian flu outbreak proved devastating to Minnesota. The virus killed more than 9 million of the state’s turkeys and chickens and wiped out the flocks of 108 poultry farms, according to a Minnesota Public Radio report.

This fall, University of Minnesota students with expertise in computer science and food security will explore new ways to curb the effects of avian flu outbreaks using a cutting-edge cognitive computing system. IBM’s Watson, a system that mimics how the human brain works by understanding naturally written language and learning from what it reads to discover new patterns and connections, serves as the basis for a special topics course.

The course, “Explore Watson,” is the result of a partnership between the U of M and IBM to discover research and business applications for cognitive computing.

“Cognitive computing can piece together data in ways researchers may not expect, finding connections they may not otherwise identify,” said Claudia Neuhauser, Ph.D., lead instructor of the course and director of the U’s Informatics Institute.

“Using a tool like Watson can help us not just in answering our research questions, but in figuring out which questions we should be asking,” Neuhauser said.

IBM partnered with the U of M as one of many institutions that are currently exploring research and business applications through Watson. Each partner institution has gained free remote access to the system, which is based out of hardware stationed at IBM, with the U of M focusing its study on avian flu. Neuhauser said the subject was chosen because it provides a real-world application for cognitive computing that has both regional importance to Minnesota and ongoing relevance to food security. The avian flu application will be the first use of the Watson system at the U, but more opportunities may follow. The system supplements more traditional forms of inquiry to aid researchers in forming connections in their data and drawing conclusions from it.

Cognitive computing systems help make better sense of enormous and complex sets of data, in particular, data in the form of plain, written English. Researchers start by feeding a database of articles into the system, which could include news stories, research publications and reports from government agencies. They then “train” the system by asking it questions written in plain English and selecting the best answers from the responses Watson provides. The system continually learns from the questions asked of it, allowing it to improve its future responses. The more information it has at its disposal, the more accurate and comprehensive its answers become since the machine cannot draw from any information beyond what its users give it. If there is no clear answer, the system will provide several possible answers, ranked by which it rates as most relevant to the question.

The machine’s learning ability provides a valuable feature that most computing systems don’t: context. Watson is capable of making sense of new information in light of what it already knows. For example, the system can use context clues to understand when to interpret a term like “risk assessment” through the lens of food security, where it would monitor threats to the food supply and the groups of people most likely to be affected, rather than through the lens of information technology, where it would measure which computer systems are vulnerable to potential cyberattacks. By understanding the context around the subject matter, Watson can better draw connections between separate pieces of information and find patterns that researchers may not have thought to look for.

Putting cognitive computing to work

“Explore Watson” will allow students to work hands-on with the Watson system to address the complex challenge of avian flu. Early in the semester, students will learn general concepts about cognitive computing and the avian flu outbreaks. Then, they will read a collection of about a hundred documents — including news articles and published research studies about avian flu, as well as state and federal government reports on its impact — and come up with hundreds of questions about them. For example, students may ask the system where the nearest outbreak was to a given farm, or how much time passed between the most recent outbreak and the one before it.

Finally, students will use their understanding of the system to come up with Watson-based ideas for real-world products or services that could aid those affected by avian flu and improve the security of the food supply. These approaches could include ways to help farmers, distributors, veterinary experts, state officials and others track outbreaks, model their growth and act to curb the damage they cause.

In addition to Neuhauser, the course’s instructors include Amy Kircher, director of the U’s National Center for Food Protection and Defense, who will provide expertise on protecting the food supply from threats like avian flu; and Mike Steinbach, Ph.D., research associate in computer science and engineering with the U’s College of Science and Engineering, who will assist with the technical aspects of using Watson. While Watson has a built-in user interface and generally does not require in-depth technical knowledge to use, it does require information to be in a specific format, meaning students will need to pare away advertisements and other content from any news stories or journal articles they input.

Going forward, Neuhauser said the emerging field of cognitive computing holds great potential for researchers to address complex challenges. Recent university courses using Watson have resulted in mobile apps that work with the system, giving users an easy and portable method of asking the questions and receiving an immediate answer. In the case of avian flu, that could mean poultry farmers would have an easy-to-access resource for information that could help them protect their flocks and manage an outbreak.

For now, however, Neuhauser and the students will focus on getting the system up and running and learning how to use it for avian flu research.

Register for “Explore Watson”

The Explore Watson course is now open for Fall 2015, with applications accepted until the start of the semester or until the course reaches capacity. Students with expertise and/or interest in one or more of the following areas are encouraged to apply: natural language processing, machine learning, cognitive computing, knowledge representation and food security.

To apply, send a statement to neuha001@umn.edu describing your interest in the topic and your expertise that could contribute to the course.