Brian Kulis is a computer scientist known for his contributions to machine learning, with a focus on metric learning, clustering, and scalable optimization algorithms. He is an Associate Professor in the Department of Computer Science at Boston University and also holds a position as a Research Scientist at Google. His work bridges theoretical foundations and practical applications, addressing challenges in high-dimensional data analysis and unsupervised learning.
Kulis's research has influenced areas such as computer vision, natural language processing, and bioinformatics, where learning appropriate distance functions is critical. He has published extensively in top-tier conferences and journals, and his work on metric learning has been widely cited. His contributions include both foundational algorithms and efficient implementations that scale to large datasets.
Early Life and Education
Brian Kulis received his Bachelor of Science degree in Computer Science from Cornell University in 2003. He then pursued graduate studies at the University of Texas at Austin, where he earned his Master of Science degree in 2006 and his Doctor of Philosophy in Computer Science in 2010. His doctoral advisor was Inderjit Dhillon, a prominent researcher in machine learning and data mining. During his PhD, Kulis focused on developing efficient algorithms for metric learning and clustering, laying the groundwork for his later career.
Academic Career
After completing his PhD, Kulis joined the Department of Electrical Engineering and Computer Sciences at the University of California, Berkeley, as a postdoctoral researcher, working with Michael Jordan. He then moved to Ohio State University in 2011, where he was an Assistant Professor in the Department of Computer Science and Engineering. In 2016, he joined Boston University as an Associate Professor, where he leads a research group in machine learning.
At Boston University, Kulis teaches courses on machine learning and data mining, and he mentors PhD students and postdoctoral researchers. He is affiliated with the Hariri Institute for Computing and the Faculty of Computing & Data Sciences. His academic service includes serving on program committees for major conferences such as NeurIPS, ICML, and CVPR, and he has been an area chair for several of these venues.
Research Contributions
Metric Learning
Kulis is best known for his work on metric learning, which involves learning a distance function from data to improve the performance of algorithms like k-nearest neighbors and clustering. His early work introduced efficient algorithms for learning Mahalanobis distance metrics, which are parameterized by a positive semi-definite matrix. He developed methods that scale to large datasets by using low-rank approximations and stochastic optimization.
One of his influential papers, "Learning Low-Rank Kernel Matrices," co-authored with Dhillon, proposed a framework for learning low-rank kernel matrices, which can be seen as a generalization of metric learning. This work provided theoretical guarantees and practical algorithms that have been adopted in various applications.
Clustering and Scalable Algorithms
Kulis has also made significant contributions to clustering, particularly in developing scalable algorithms for large-scale data. He worked on spectral clustering and its approximations, including the use of Nyström methods to reduce computational complexity. His research on online and streaming clustering has addressed the challenge of processing data that arrives sequentially, which is relevant in real-time applications.
In collaboration with colleagues, Kulis developed algorithms for subspace clustering, which aim to group data points that lie in low-dimensional subspaces. This has applications in motion segmentation and face recognition. His work often emphasizes theoretical guarantees, such as convergence rates and approximation bounds, alongside practical efficiency.
Deep Learning and Applications
More recently, Kulis has explored connections between metric learning and deep learning. He has investigated how neural networks can be trained to learn embeddings that preserve similarity relationships, which is useful for tasks like face verification and image retrieval. His research has also touched on unsupervised and semi-supervised learning, where labeled data is scarce.
Kulis has collaborated with researchers in computer vision and natural language processing, applying his methods to problems such as object recognition and document clustering. His work has been supported by grants from the National Science Foundation and other agencies.
Selected Publications
Kulis has authored or co-authored over 60 peer-reviewed papers. Some of his most cited works include:
- "Learning Low-Rank Kernel Matrices" (2006, with Inderjit Dhillon)
- "Metric Learning: A Survey" (2012, with a review of the field)
- "Scalable Clustering: Algorithms and Applications" (2013, with colleagues)
- "Deep Metric Learning: A Survey" (2019, with co-authors)
His papers have appeared in journals such as the Journal of Machine Learning Research and IEEE Transactions on Pattern Analysis and Machine Intelligence, as well as in proceedings of NeurIPS, ICML, and CVPR.
Awards and Honors
Kulis has received several awards for his research and teaching. He was a recipient of the National Science Foundation CAREER Award in 2014, which supports early-career faculty. He has also been recognized with best paper awards or nominations at conferences, including a notable paper at the Conference on Uncertainty in Artificial Intelligence. In 2020, he was named a Senior Member of the Association for the Advancement of Artificial Intelligence (AAAI).
Industry Experience
In addition to his academic role, Kulis has worked in industry. He has been a Research Scientist at Google since 2017, where he works on machine learning problems related to large-scale systems. This role allows him to apply his research to real-world products and collaborate with engineers. His industry work has included projects on clustering and similarity search, which are relevant to search and recommendation systems.
Teaching and Mentoring
Kulis is known for his dedication to teaching and mentoring. He has developed graduate-level courses on machine learning that emphasize both theory and practice. He has supervised numerous PhD students who have gone on to academic and industry positions. His mentoring style encourages students to tackle challenging problems and to communicate their findings clearly.
Impact and Legacy
Brian Kulis's work has had a lasting impact on the field of machine learning. His research on metric learning has provided a foundation for many subsequent developments, and his scalable algorithms have been used in a variety of domains. He is frequently invited to give talks at conferences and workshops, and his opinions are sought on topics related to unsupervised learning and optimization.
As of the mid-2020s, Kulis continues to be an active researcher, publishing new work and collaborating with colleagues across institutions. His contributions have helped shape the way machine learning models learn from data, particularly when labeled examples are limited.
See Also
- Machine learning
- Artificial intelligence
- Deep learning
- Neural network
- clustering (not in list, but related)
References
This article is based on publicly available information about Brian Kulis's career and publications. Specific details about his life and work are drawn from his academic profile and research papers.