Valentin Robu

I am a third-year PhD student at CWI, Dutch National Research Center for Mathematics and Computer Science, Amsterdam.

My research mostly deals with simulation of electronic markets as coordination mechanisms between autonomous, self-interested agents. There are two main perspectives to be considered in this research: 1. modeling the structure of the market itself, such as to insure some desirable properties (e.g. incentive compatibility, revenue equivalence etc.), 2. modeling the preferences, risk attitude, learning and adaptation of the participating agents themselves and examining (through simulations) the effect they have, both on individual agent outcomes and the reached market equilibrium. In my work, I mostly take the second approach. In order to model agent learning, I use a variety of techniques including probabilistic learning, evolutionary computation, RL etc. If you want to know more about it, here is my homepage. Regarding to my non-scientific life (yep, there is one!), I enjoy sports, reading, listening to good music, going out, travelling to exotic places (well, if the budget permits).

And here are the answers to Dan's questions:

> 1. What topics do you have some expertise in and would you be willing to help others learn them?

My research has mostly focused on computational aspects of multi-agent systems, in particular on using machine learning techniques (e.g. utility graphs, Bayesian reasoning) to implement agent reasoning electronic market situations. My expertise in the area of design of automated negotiation strategies and protocols and design of auction-bidding strategies. If other participants are interested in the topic, I would be glad to talk to them.

> 2. What do you want to learn?

Basically, the topics covered in week 3 of the course are the closest to my own line of work. I am especially interested to hear about current work performed at SFI on artificial market simulations, evolutionary and econo-physics models of market behavior, as well as the topic of modeling the dynamics of complex networks.

Topics in neurocomputing and cognitive science also seem interesting. So far in my research, I mostly dealed with MACHINE learning and decision making - but I would find it interesting to also get more in-depth perspective on how humans performs these tasks. I have little of a formal background in some topics, such as biology or ecology, but I'm always interested to learn more.

> 3. Do you have any projects that would benefit from interdisciplinary approach?

I am currently working in the framework of a project called DEAL (Distributed Engine for Advanced Logistics), whose goal is to improve efficiency of complexy suply chain and networks, consisting of self interested agents. My own research in this project concern mainly the mechanisms task allocation and market-based allocation of loads. Having an inter-disciplinary perspective is important - I am particular interested to speak to people working in areas such as OR, economics or cognitive science.

> 4. Do you have any ideas for what sort of project you would like to attack this summer?

I do not have a definite idea yet, but I was thinking something on the line of artificial market simulations (I have seen quite a few interesting papers from Santa Fe faculty on this topic). Otherwise any topic involving simulation of complex economic networks, logistic networks - using evolutionary or other agent-based models would be of interest. Of course, the exact choice of topic depends on the faculty/other participants.

> 5. What's your favorite "big problem"?

My favourite big problem is probably how to design a truly intelligent machines - one that can replace humans in a wide variety of settings. Since getting a machine that would pass the Turing test is probably too far-fetched (now and in the future), at least getting an agent that performs indistuinguishly well from a human in a specific domain (for example, economic decision making) would still be something.

> 6. If you were given the opportunity to see where we were in one hundred years with respect to progress on one problem/subject, what would it be?

I would probably ask to what extent have agents replaced humans in different areas (for example, will there be many human logistic planners, stock brokers etc. ?). Next, I would ask what type of machine learning techniques they are using to make them work.