Terran Lane, Assistant Professor

Computer Science Dept., UNM

I'm an assistant professor of computer science at UNM. My primary (academic) interests are: machine learning; reinforcement learning, behavior, and control; and artificial intelligence in general. I'm also interested in computer/information security/privacy and bioinformatics.

Publications list


Research

Machine learning is simultaneously a pragmatic discipline, concerned with the analysis of complex data from a variety of fields, and a theoretical one, concerned with the principles of what is learnable, how to represent acquired knowledge, how to deal with complexity/dimensionality, the interactions between learned knowledge and behavior, how to measure acquired knowledge, and so on. The tools we use include statistics, algorithms, knowledge representation, database theory, linear algebra, and (in recent developments) topology.

My personal research interests include behavioral modeling and learning to act/behave (i.e., reinforcement learning), scalability, representation, and the tradeoff between stochastic and deterministic modeling. All of these represent different facets of my overall interest in scaling learning methods to large, complex spaces and using them to learn to perform lengthy, complicated tasks and to generalize over behaviors. While I attempt to understand the core learning issues involved, I often situate my work in domain studies in practical (well, ok, semi-practical anyway) problems. Doing so both elucidates important issues and problems for the learning community and provides useful techniques to other disciplines.

Scaling Techniques for Planning and Learning in MDPs and POMDPs
The reinforcement learning paradigm rests on the foundation of the theory of Markov decision processes (MDPs) and their bigger, badder cousins, partially observable MDPs (POMDPs). While tractable methods for optimal planning in small MDPs have been understood for decades now, we still hit a wall when we try to scale to larger domains. In this project, I'm working on techniques for performing approximate planning and learning in large (e.g., 2^500 states or more) models.

Anomaly detection for computer security

A number of critical problems in computer security can be viewed as distinguishing some "normal" circumstance from "anomalous" or "abnormal" circumstances. For example, we can think of computer viruses as being (syntactic and begavioral) abnormal modifications to normal programs.

Computational modeling of RNAi

RNAi web site, with off-target tools

RNA interference is a recently discovered biological mechanism that appears to be a widespread and highly evolutionarily conserved (i.e., ancient) genetic immune mechanism. Research in the past five years or so has shown that it is involved in defense against some classes of viruses and transposons, as well as in certain cellular regulatory mechanisms. The exciting feature of this mechanism is that it can be exploited mechanistically to target some viruses, offering hints of the first possible direct treatment for viral infections, as well as to selectively knock down the expression of specific genes (via posttranscriptional disruption of the corresponding mRNA), greatly simplifying gene function studies.

Unfortunately, while a reasonable qualitative picture of the mechanics of RNAi has emerged, we are still far from a quantitative and predictive understanding. Currently, activating sequences (siRNA or dsRNA) are hand-picked employing rough "rules of thumb". Our group is attempting to build more quantitative and predictive models by applying machine learning-based bioinformatic techniques to genome and RNAi data sets. Our goals are to produce high-accuracy predictions of the activity of specific sequences and, hopefully, to shed light on some of the mechanical and evolutionary details of RNAi. Along the way, we hope to answer pragmatic questions such as the expected false positive rate (i.e., rate of knockdown of untargeted genes) and minimal covering sets for gene families.

Students and Postdocs

I'm fortunate enough to have the chance to work with a number of excellent students here at UNM who're deeply involved in various interesting ML projects including analysis of fMRI (neuroimaging) data, bioinformatics, and reinforcement learning. Check in with each of them to find out what they're up to!
  • Blake Anderson (MS): Analysis of fMRI data.
  • Eva Besada-Portas (visiting scholar): Multi-agent reinforcement learning and control.
  • Patrick Carlson (undegrad): Reinforcement learning for malware detection.
  • Nathan Fabian (MS): User modeling and behavioral cloning for interactive games.
  • Laura Glendenning (undergrad): Reinforcement learning; psychologically grounded ML.
  • Will McMain (undergrad): Reinforcement learning for malware detection.
  • Diane Oyen (MS): Analysis of fMRI data.
  • Dennis Paiz-Ramirez (undergrad): Reinforcement learning for malware detection.
  • Sergey Plis (Postdoc): Bayesian MEG/fMRI data fusion.
  • Sushmita Roy (PhD): Genomics and bioinformatics.
  • Mark Scully (MS): Network inference from neuroimaging data.
  • Avani Wildani (MS): Data mining from neuroinformatics and psychological data.
  • Ben Yackley (MS): Network inference from neuroimaging data.

Classes

More past classes

Alumni

Here are some of the students who have graduated from my lab and gone on to (hopefully illustrious) endeavors elsewhere:

Undergraduate Research Positions

The UNM Machine Learning group has a number of positions open for undergraduate researchers. If you're considering grad school or would just like to get beyond your classes and into the cutting edge of the research world, please get in touch with me.

Embedded Machine Learning Systems
In collaboration with the University of Oklahoma, students will have the opportunity to learn about and work on machine learning in systems ranging from robotics to weather prediction to computer security. Students will have the opportunity to interact with colleagues at OU, participate in the broader research community, and travel to cutting edge research conferences.