Abram Friesen
afriesen at cs dot washington dot edu
Room 609, Paul Allen Center
University of Washington

I am a Computer Science and Engineering Ph.D. student at the University of Washington, where I work with Pedro Domingos. I received my B. Eng. in Computer Engineering from the University of Victoria in British Columbia, Canada.

Nonconvex optimization problems are prevalent in both the scientific and machine learning communities. My current research is on developing a novel algorithm for nonconvex optimization that uses techniques from the combinatorial optimization literature to reduce the size of the search space by an exponential factor. We hope to use this algorithm to advance the state of the art in protein folding, one of the most important problems in biochemistry. Similarly, a significant advance in nonconvex optimization could drastically improve machine learning by redefining the types and complexity of models and algorithms that we can feasibly work with. Longer term, I am interested in developing general learning techniques for artificial intelligence that are domain agnostic and rely on fewer assumptions than modern methods.

In previous work, I have explored nonparametric Bayesian modeling, planning and decision making, reinforcement learning, imitation, and goal inference. I've had the pleasure of being advised by both Rajesh Rao and David Wingate, and have had the good fortune to visit other research labs, including MIT in the labs of Leslie Kaelbling and Josh Tenenbaum, Berkeley with Tom Griffiths in the Computational Cognitive Science Lab, and an internship at Intel Labs in Seattle, working with Dieter Fox on Human Robot Interaction.

Here is a link to my (slightly out of date) CV.

Projects and Publications

Planning and Decision Making
We present a normative model of decision making that incorporates prior knowledge in a principled way. We show that experimentally-observed behaviours emerge naturally when decision making is viewed within the framework of partially observable Markov decision processes (POMDPs), the basis of our model.

Yanping Huang, Abram L. Friesen, Timothy D. Hanks, Michael N. Shadlen, and Rajesh P. N. Rao (2012). How prior probability influences decision making: a unifying probabilistic model. In Proceedings of the 26th Annual Conference on Neural Information Processing Systems. Lake Tahoe, Nevada. December, 2012. (pdf)

Inferring Object Reference Frames
We propose an ideal observer model based on nonparametric Bayesian statistics for inferring the number of reference frames in a scene and their parameters.

Joseph L. Austerweil, Abram L. Friesen, and Thomas L. Griffiths (2011). An ideal observer model for identifying the reference frame of objects. In Proceedings of the 25th Annual Conference on Neural Information Processing Systems. Granada, Spain. December, 2011. (pdf)

Goal Inference
We introduce a Bayesian model where gaze-following occurs as a consequence of goal inference in a learned probabilistic model. We use Gaussian processes to provide a nonparametric and flexible approach for learning the policy and transition functions.

Abram L. Friesen and Rajesh P. N. Rao (2011). Gaze Following as Goal Inference: A Bayesian Model. In L. Carlson, C. Hölscher, & T. Shipley (Eds.), Proceedings of the 33rd Annual Conference of the Cognitive Science Society. Boston, MA: Cognitive Science Society. July, 2011. (pdf)

Imitation Learning
By combining the flexibility of imitation learning with the temporal abstraction provided by hierarchical actions we are able to demonstrate significant performance and scalability improvements in an artificial domain modeled after a cognitive science puzzle-box task.

Abram L. Friesen and Rajesh P. N. Rao (2010). Imitation Learning with Hierarchical Actions. In Proceedings of the International Conference on Learning and Development (ICDL), Ann Arbor, MI, USA. August, 2010. (pdf)

Query Progress Estimation
In this paper, we introduce Parallax, the first non-trivial progress estimator for parallel MapReduce queries.

Kristi Morton, Abram Friesen, Magdalena Balazinska, and Dan Grossman (2010). Estimating the Progress of MapReduce Pipelines. In Proceedings of the 26th IEEE International Conference on Data Engineering (ICDE), Long Beach, CA, USA. March, 2010. (pdf)