Terran Lane
Location: (Albuquerque, New Mexico)
Personal Research Web Page: http://www.cs.unm.edu/~terran/
Keywords: Machine learning, scientific data mining, neuroscience, bioinformatics, interdisciplinary data analysis, networks, graphs, topology, applications, statistical machine learning, geometry, complexity
Posted on: Monday, May 3rd, 2010
Broad Research Area: AI / Machine Learning / Robotics / Vision, Databases / Information Retrieval / Data Mining, Other, Scientific/Medical Informatics
Research Interests:
I am interested in machine learning and data mining for complex and large-scale scientific data, motivated by real-world application data and close collaborations with domain scientists. Recently, I am most active in ML problems related to networks — structure inference, prediction of network function or behavior, network dynamics, and so on. These studies have motivated me to examine the intersection of ML, statistics, and topology. But I also have been interested in decision theory (reinforcement learning and planning), user modeling, representation, and relational learning. I am not a theorist, but I do believe in the importance of mathematical foundations in the design of learning algorithms.
On the application front, I feel that many, if not most, important problems in ML have been motivated by interactions with new, complicated forms of data. I enjoy collaborations with scientists in disciplines including genomics, immunology, neuroscience, geology, computational chemistry, computer security, and behavior modeling. My goal is to help these groups advance their own fields, while developing novel ML methods and methodologies.
For more information on my research program, please check out my web page:
http://www.cs.unm.edu/~terran/
I have also recently become interested in the emergence and growth of complexity in systems like evolution or economics. While these systems bear some similarity to learning systems (in the sense of extracting information from the environment, encoding/compressing it, and using the new representation to improve performance in the future), they also exhibit important behaviors beyond those of traditional machine learning systems. For example, while all ML algorithms that I know of hit some eventual performance asymptote, these “natural learning systems” never have. I am interested in exploring the ties between these phenomena, toward developing unbounded learning systems.
