Belief and Truth in Hypothesised Behaviours

My PhD student, Stefano Albrecht, will have his viva voce examination this Wednesday. As is the convention in some parts of our School, he will give a pre-viva talk at IF 2.33 between 10 – 11 am on Wednesday, 19th August.

His talk abstract: This thesis is concerned with a specific class of multiagent interaction problems, called ‘ad hoc coordination problems’, wherein the goal is to design an autonomous agent which can achieve flexible and efficient interaction with other agents whose behaviours are unknown. This problem is relevant for a number of applications, such as adaptive user interfaces, electronic trading markets, and robotic elderly care. A useful method of interaction in such problems is to hypothesise a set of possible behaviours, or ‘types’, and to plan our own actions with respect to those types which we believe are most likely, given the observed actions. We investigate the potential and limitations of this method in the context of ad hoc coordination, by addressing a spectrum of questions pertaining to the evolution and impact of beliefs as well as the implications and detection of incorrect hypothesised types. Specifically, how can evidence (i.e. observed actions) be incorporated into beliefs and under what conditions will the resulting beliefs be correct? What impact do prior beliefs (before observing any actions) have on our ability to maximise payoffs in the long-term and can they be computed automatically? Furthermore, what relation must the hypothesised types have to the true types in order for us to solve our task optimally, despite inaccuracies in hypothesised types? Finally, how can we ascertain the correctness of hypothesised types during the interaction, without knowledge of the true types? The talk will conclude with interesting open questions and future work.

While his thesis will become available in due course, you can get an idea of the main argument in this submission to the AI Journal: http://arxiv.org/abs/1507.07688, entitled Belief and Truth in Hypothesised Behaviours.

The deal with the devil

In many of my papers, we have been inspired by and found use for geometric methods, as have many other roboticists. One reason I like geometry is the naturalness of that language for describing properties of dynamics – again something physicists have long known and roboticists have likewise come to understand. However, a constant frustration has been the difficulty of efficiently computing with geometric objects, especially when the focus is on abstraction (e.g., global properties of manifolds) rather than, say, efficient location of points in a planar region. In our most recent paper, at R:SS 2014, we found that the computational barriers begin to come down when you bring in more tools from algebra – we used matrix computations based on an algebraic topological formulation of what we were after, and that became quite efficient.

In this context, I found the following quote very interesting:

Algebra is the offer made by the devil to the mathematician. The devil says: “I will give you this powerful machine, it will answer any question you like. All you need to do is giving me your soul: give up geometry and you will have this marvellous machine.”

– M. Atiyah, “Mathematics in the 20th century,” in Mathematical Evolutions. Providence, RI: Mathematical Association of America, 2002.

On disciplinary boundaries

A poet once said, “The whole universe is in a glass of wine.” We will probably never know in what sense he meant that, for poets do not write to be understood… How vivid is the claret, pressing its existence into the consciousness that watches it! If our small minds, for some convenience, divide this glass of wine, this universe, into parts– physics, biology, geology, astronomy, psychology, and so on– remember that nature does not know it! So let us put it all back together, not forgetting ultimately what it is for. Let it give us one more final pleasure: drink it and forget it all!

-Richard Feynman, from Six Easy Pieces

RoboCup 2011

The past month has been hectic, consumed by our debut at RoboCup 2011 in sunny Istanbul, Turkey. It was a very interesting and educational experience for the team and me. We have received an unexpected amount of publicity (http://wcms.inf.ed.ac.uk/ipab/robocup/new-and-events/media-coverage) which makes me a bit nervous – this was just our very first year in a competition that hosts many large and established teams with years of experience!

The summary of our performance:

In the Standard Platform league, we got eliminated in the group stages 0-1, 0-3,0-1 but that doesn’t say the whole story. In both the 0-1 , matches, we lost due to our own mistakes and our level of play was quite evenly matched against our opponents!

We had more success in our matches within the 2D simulation league. The virtual team, mainly developed by Majd Hawasly, won three matches in the group stages but then lost to more established teams when we got further (details here: http://sourceforge.net/apps/mediawiki/sserver/index.php?title=RoboCup2011/Competition).

In both cases, we learnt a lot. On the positive side, it was clear that we fully understood the basic technical problems but needed much more time and, importantly, manpower to develop our core technology. Our robots were performing on par with other young teams when they worked well but the robustness needs improvement. However, we also learnt that we weren’t yet fully match ready. We need to play a lot more practice matches to fine tune our technology before the next world cup in Mexico.

If you want to watch one of our matches, follow the links below. These videos were shot and uploaded by our opponents, the Germans (Nao Devils – the team that played for the cup in the finals):

http://www.youtube.com/watch?v=HB5RKTJnmWY (First half: Edinferno is in Red)

http://www.youtube.com/watch?v=nlw7EUw9U9A (Second half: Edinferno is in Blue)

The people in our Standard Platform League team (Stathis Vafeias, Aris Valtazanos, Chris Towell and me):

… and finally, the entire Standard Platform League photo (to be precise, only the human participants :-)):

CompuCast

My friends have recently started a podcast on topics in computer science, called CompuCast.

Descriptor for the latest episode:

This episode we pay tribute to the late Robin Milner, we avoid avoiding deadlocks and warn against expecting everything to be as easy as it is for physicists. We pit Edd, our producer, against three artificial intelligences to find out if he’s smarter than them. And, as always, we bring you the computer science quiz and the joke of the month (which Edd promises will not be as lame as usual!).

This episode includes a small contribution from yours truly.

Welcome

Welcome to my blog. I plan to use this space to jot down thoughts arising from my research and teaching, or perhaps on more general issues concerning a life in science. Hopefully, some of these thoughts will lead to further and deeper conversations, enabling me to learn and discover.

Let me begin by introducing myself. I am a computer scientist, originally trained as an electrical engineer, working in the area of machine learning as applied to problems involving complex dynamical systems. Much of my work centers on understanding and synthesizing complex motion – in settings ranging from humanoid robotics to bio-molecular systems. I have a particular interest in how complex motion strategies are structured, acquired and improved from continuous experience.

More broadly speaking, I am interested in mathematical and algorithmic problem solving, often drawing on foundational tools from dynamical systems, geometry, topology and probabilistic methods.