The era of big data is introducing big problems. Data grows at a rapid pace that makes it difficult for researchers to analyze. A new ISE faculty member is addressing these challenges through her research in large-scale optimization problems.
Niao He, who joined ISE faculty in the spring of 2016 as an assistant professor, said the prospect of solving real-world problems by advancing algorithmic design and understanding the theoretical complexity of algorithms is what drives her research.
“Since my childhood, I’ve always been fascinated with numbers,” He said. She studied mathematics at the University of Science and Technology of China, where she was one of the youngest college students in the country, and then earned her master’s in computational science and engineering and a Ph.D. in operations research from the Georgia Institute of Technology.
She said her decision to join ISE faculty involved the University’s reputation for innovative research and high-quality education, particularly within the Department of ISE.
“It’s a wonderful place that I can easily collaborate with people, because we have experts in all research areas,” she said. “I really enjoy working with my colleagues and students here. They are really friendly, brilliant, and most importantly, passionate about what they do.”
He’s research focuses on developing fast and reliable algorithms to solve large-scale optimization problems. Algorithms like these are the cornerstone of numerous real-world applications in machine learning, finance, computer vision, and many other fields.
One of her recent projects involved designing an efficient algorithm for services that have a recommendation system, such as Amazon, Netflix or Pandora. “Based on the history of a user’s activities — for instance, clicking a link, watching a movie, or posting a tweet — a good service system should make the right personalized recommendation to the right customer at the right moment,” He said.
Creating an effective recommendation system, He explained, requires finding the proper way to mathematically model the dynamics of user preference. Then, to make the recommendation fully automated and timely, the model has to be solved efficiently. That’s why algorithmic design plays such an important role.
He proposed modeling the user preference by taking into account how preferences may change over time. For example, what if a user wants to watch a certain genre of movie at a certain time of the day — a romantic movie in the evening, or a comedy during the day?
So, He and her fellow researchers created a time-sensitive recommendation model that captures patterns from user activities and produces a well-structured optimization problem. They also proposed a novel algorithm which is both efficient and has a theoretical basis.
Another application of large-scale optimization that He is working on is image classification. Websites such as Flickr have millions of images that need to be classified into different groups, which creates large amounts of data and, therefore, a large-scale problem. A single image could represent about one million variables that need to be considered.
The current state-of-the-art method used to classify these images is called artificial neural networks. These networks are composed of a large number of interconnected layers, similar to the human brain, and are flexible enough to estimate any function. He explained that this method works well in practice, but the reasoning for the mechanism behind it still remains a mystery among researchers.
Together with her collaborators, He developed an algorithm that differentiates from neural networks yet performs comparably in real-world datasets. And the fact that it is theoretically grounded also sets it apart.
“If you know the theory behind it, you know how to improve it. You know how to move forward, and you know in what way it may be connected to solving other problems,” she said.
Providing theory behind solutions has become a central theme of He’s work. It’s something she thinks should be a critical aspect of all research.
“People in industry care more about performance. As long as an algorithm works fine, then they are going to adopt it,” she said. “But as a researcher, we need to truly understand what’s going on.”
She said her ultimate career goal is to further bridge the gaps between modern optimization theory and new applications in today’s big-data environment. Bringing the theoretical background to reinforce solutions, she said, will be beneficial for a wide range of real-world problems.
“There are still lots of open questions and challenges there, which is also why this field is so intriguing and actively explored by thousands of mathematicians and computer scientists around the world.”