Putting the magic in the machine since 1980.

Monday, November 19, 2007

The Internet, Growth, and Students

Today I gave a 1-hour talk to our freshmen on the history an future of the Internet. Obviously an impossible task but I did my best. I hope I conveyed to them the endless possibiblities that the Internet has opened up.

In preparation for this talk I have been asking our students: how much do you think the .com bubble affected the growth of the Internet? The general belief is overwhelmingly clear. Students believe the bubble had a large impact on the Internet growth. Of course, this is completely wrong. Check out my chart:

Nasdaq vs. Number of Internet Hosts vs. Number of Websites

The bubble had absolutely no impact on the growth of the Internet which continues to grow exponentially, doubling every three years. The number of websites is doubling at least every two years. The implications of these facts are mindboggling! But, the general public, even our students, still seems to feel that software is done. Oh well, it just means more money for those of us who can program!

Tuesday, October 23, 2007

Combinatorial Auction Model

One can visualize a combinatorial auction (with OR bids only) as a graph with two types of nodes: bids and items. Each bid node is labelled with the price of the bid and connected to all the items it is over. This is the visualization I have implemented in my combinatorial auction NetLogo model. It implements the branch-and-bounds algorithm on the branch-on-bids tree as found in my textbook.

Thursday, October 11, 2007

Iterated Equiresistance Model

A couple of years ago I built this simple NetLogo model that implements the iterated equiresistance algorithm in randomly generated exchange networks, as used by network exchange theory in Sociology. It now looks kinda dated and needs to be cleaned up but I thought I'd post it nonetheless. I'm using it in class as a demo.

Wednesday, October 10, 2007

Pareto Learning Model

I have posted a new netlogo model on pareto learning which implements the algorithms from this paper. This is a quick and dirty implementation for an in class demo. As nearly always happens when I do these, I found a slight problem in the paper. They specify two different ways in which the agents choose an action. Namely, first they say that the agents choose actions stochastically based on their expected utilities, then they say that they choose their best action and with probability epsilon choose a random action. After implementing both strategies it became clear that the second one is the one they actually used. Clearly, this was just a problem of the prose being a bit confusing (at least, for me).

It is also interesting to note how such a small change in the action choice method can have such large effects on the system's behavior. Because of this I have to say that the results from this model are not stable, thus they are not of deep significance.

Monday, August 13, 2007

PhD Graduation Pictures

The PhD hooding ceremony was this weekend. Hrishi, Hong and Karthik took pictures, some of which I have shamelessly borrowed and placed on my own flickr set for all to see.

Thursday, August 2, 2007

Jiang Receives PhD

The other Phd graduate this summer is Hong Jiang and her thesis is

  • Hong Jiang. From Rational to Emotional Agents. University of South Carolina, 2007.
    To date, most research on multiagent systems has focused on rational utilitymaximizing agents. However, theories show that emotions have a strong effect on human~s physical states, motivations, beliefs, and desires. The details have not been explicated clearly so far. In artificial intelligence, emotions have begun to receive more attention, but mostly in human-robot/computer interaction. The research on applying emotions to agents~ decision-making is still very limited. Can agents be intelligent without emotions? We believe that, whether for humanlike or non-human-like agents, the effect of emotions on decision-making cannot be ignored, since agents with high emotional quotients (EQs) can be built to have better performance in complex dynamic environments than purely rational agents. This research focuses on the effects of emotions on decision-making. Taking into account the incompleteness of emotion theories and emotional differences among individuals, I describe EBDI, a common architecture for emotional agents, which specifies a separate emotion mechanism within an agent, instead of trying to model emotion mechanisms to reflect the reasoning process specifically, like most researchers have done. It reflects the practical reasoning process, and one can select and apply part of an emotion theory into the architecture as needed. Sample agents in Tileworld are presented and the results show that an EBDI agent can have better performance than traditional BDI agents. To apply EBDI in negotiation, a plug-in is designed, which modifies the OCC model, a standard model for emotion synthesis, to generate emotions. Considering the possibility of incorporating emotions into negotiation, I generate EWOD (EmotionalWorth- Oriented Domain), which requires numerical emotions. Thus, a mapping from 22 OCC emotions to 3-dimension numerical PAD emotions is given. Finally, I describe how PAD emotions affect the negotiation strategy and provide an evaluation which shows that it can be used to implement emotional agents that mimic human emotions during negotiation. Thus we can design high EQ agents for negotiation according to specific design purposes. Since negotiation is used widely in many different domains, this research, based on a general process of negotiation, can also be widely applied to other areas.

Congratulations are also in order for Hong. Her research is highly innovative, crossing boundaries between computer science, cognitive science, and Sociology. She has shown how simulated emotions can be incorporated into negotiating agents for the betterment of the whole agent society, as well as how agents can be made to behave like normal irrational humans. Check our her extensive list of papers. She will be joining the faculty at Benedict College.

Goradia Receives PhD

We have two new PhD graduates this summer. One of them is Hrishikesh Goradia who has successfully defended his PhD thesis:

  • Hrishikesh J. Goradia. Automated Negotiation Among Autonomous Agents in Negotiation Networks. University of South Carolina, 2007.
    Distributed software systems are a norm in today~s computing environment. These systems typically comprise of many autonomous components that interact with each other and negotiate to accomplish joint tasks. Today, we can integrate potentially disparate components such that they act coherently by coordinating their actions via message exchanges. Once this integration issue is resolved, the next big challenge in computing is the automation of the negotiation process between the various system components. In this dissertation, we address this automated negotiation problem in environments where there is a conflict of interest among the system components. We present our negotiation model - a negotiation network - where a software system is a network of agents representing individual components in the system. We analyze the software system as a characteristic form game, one of many concepts in this dissertation borrowed from game theory. The agents in our model preserve the selfinterest of the components they represent (their owners), and make decisions that maximize the expected utilities of their owners. These agents accomplish joint tasks by forming coalitions. We show that the problem of computing the optimal solution, where the utilities of all agents are maximized, is hyper-exponential in complexity. We present an approximate algorithm for this hard problem, and evaluate it empirically. The simulation results show that our algorithm has many desirable properties - it is distributed, efficient, stable, scalable, and simple. Our algorithm produces the optimal (social welfare maximizing) solution for 96% of cases, generates maximal global revenue for 97% of cases, converges to 90% of the best found allocation after only 10 rounds of negotiation, and finds a core-stable solution for revenue distribution among the agents for cases with nonempty core. Finally, to ensure stability for all cases, we present a sliding-window algorithm that computes the nucleolus-stable solution under all situations.

Congratulations are in order for Hrishi! His thesis presents some highly innovative algorithms for automated negotiation, a topic that I expect will see many applications in the near future as semantic we technologies come of age and automated service orchestration becomes the norm in business processes. Go read it! I also want to point out that Hrishi's research contributions go far beyond his thesis; many of Hrishi's publications did not even make it into his thesis. He has accepted a position in the computer science department at Western Carolina University. We wish him well.

Tuesday, March 27, 2007

Supply Chains, Terrorists Attacks, and Agent-Based Modeling

I have recently been doing some work on supply chains and, more generally, agent-based modeling in general. There is a certain art to making an agent-based model that is complex enough to capture the problem you are trying to model but still simple enough to remain manageable. Languages like NetLogo are excellent in that they eliminate nearly all extraneous code and one is left with something that almost (OK, maybe I'm being generous) looks like pseudo-code. Nearly all the code relates directly to the problem.

Still, its lots of fun.

On that note, I have posted a NetLogo model on supply chain survivability. The model itself is interesting in that I had to implement Dijistra's to calculate the minimal path between every pair of nodes. This was needed to later calculate the graph's clustering coefficient and the characteristic path length—two measures of a graph's connectivity, the smaller they are the closer everyone is to everyone else (thus, the shorter it takes to get items to the customers). Also note that the model generates small-world graphs (using preferential attachment) as well as random graphs. This model is the basis of some work we are doing. Anyone out there interested in this kind of stuff, let me know.

Friday, February 2, 2007

Three New Papers for AAMAS

We had three papers accepted to this year's Autonomous Agents and Multiagent Systems Conference. The first one will get a full presentation and the other two will be presented as posters.

  • Benito Mendoza and José M. Vidal. Bidding Algorithms for a Distributed Combinatorial Auction. In Proceedings of the Autonomous Agents and Multi-Agent Systems Conference, 2007.

    Benito builds on my previous effort to design a bidding algorithm for the PAUSE auction by providing a much faster bidding algorithm. The PAUSE auction is a combinatorial auction in which the bidders themselves solve the winner determination problem in order to win. They do so because that is the only way to win! Thus, the auction provides a simple way to distribute the computation among the agents. Our algorithms implement the myopically-optimal strategy for the agents in such an auction which is to calculate a new winning bidsest but only if in that bidset they would get more utility than they get from the currently winning bidset. In other words, if I don't get any more utility from a new bidset then I won't bother to calculate one (of course, some else might think differently.

  • Hrishikesh J. Goradia and José M. Vidal. An Equal Excess Negotiation Algorithm for Coalition Formation. In Proceedings of the Autonomous Agents and Multi-Agent Systems Conference, 2007.

    Hrishikesh's attacks a problem similar to the one solved by the PAUSE auction but in this case it is the goods (items for sale) which negotiate to find the best allocation. This maps better to service-oriented architecture scenarios where various service providers are trying to sell their services but the buyers only want bundles. For example, you might only want a shipping service if you can also buy the book you want, both for less than $10. We present an algorithm based on Equal Excess theory which is an old solution concept to the negotiation problem (aka characteristic form game).

  • Hong Jiang, José M. Vidal, and Michael N. Huhns. EBDI: An Architecture for Emotional Agents. In Proceedings of the Autonomous Agents and Multi-Agent Systems Conference, 2007.

    Hong is studying the problem of how to incorporate established emotional models from psychology into autonomous agents, namely BDI agents. This paper presents a first try at an architecture that will blend these two. The goal is to develop agents that behave like humans—with all the irrationality that that implies. We speculate that as agents start to take over more of our tasks, such as buying/bargaining with others, that users will be angry if their agents do not behave like they would, even if such a behaviors would be deemed irrational by a standard utility-maximizing model.