Subramanian Ramamoorthy

Assistant Professor, Sch. of Informatics, U. of Edinburgh

Abstract Learning: "The Transfer Problem"
21 minutes, 9.6mb, recorded 2011-11-10
Subramanian Ramamoorthy

The "problem of transfer learning," is the fundamental problem of knowing how to apply what you know (wisdom vs. knowledge). Prof. Ramamoorthy studies how highly autonomous agents can learn to transfer their skills to new situations. Ramamoorthy's research considers the similarity of situations in an abstract sense. For example, once agents have learned to consider the possibility that information from other agents may be deceptive, that's obviously a huge step. Then, to transfer that learning, agents need to spot new situations where deception may be an issue.

"I prefer a few rich interactions in small contained settings," Prof. Ramamoorthy explains, comparing his ideal to the relationship between a farmer and his domestic farm animals, where each animal interacts with a few others, and is useful in specific rich ways. Each actor develops sophisticated relationships with a limited set of others. In other robotics labs, it is popular to study the intelligence of ensembles, like ants or bees, where agents have very limited autonomy. "That's not my focus," he emphasizes. Ramamoorthy tries to constrain his environment and empower a lmiited number of agents. His goal is to develop theories about how agents can apply skills in novel situations.


Subramanian Ramamoorthy is an Assistant Professor at the School of Informatics at the University of Edinburgh. His research explores “problems of autonomous decision making over time by agents interacting with complex dynamical systems, motivated by applications in robotics and electronic markets, addressed using a combination of techniques from machine learning and mathematical systems theory.” Previously Prof. Ramamoorthy was with the University of Texas at Austin, where he received a PhD in Electrical and Computer Engineering from The University of Texas at Austin. Prior to that, he was at National Instruments Corp. in the control design, dynamic simulation and computer vision groups.

Resources

This free podcast is from our Flexible Elements with Per Sjöborg series.

For The Conversations Network:

Photo: Subramanian Ramamoorthy