Miguel Rebollo

On the design of mWater: a case study for Agreement Technologies
Pablo Noriega

About how to deal with water problems, involving many stakeholder with conflicting interests. This problem is modelled using electronic institutions. My comment: I think that is importan to consider more general works. I particularly like the papers about complex social ecological systems (Ostrom 2009 and Meizen-Dick 2007 ) which combines the environment, the resources, the users and the institutions and their relationships in a complete model. eInstitutions are very useful to model the institutional part (administrations, goverment…), but a more open environment (Thomas) is needed to take into account individual users and their social relationships to create a self-adaptive subsystem. I guess that this is the correct direction to deal with this problem. The vision that Carlos, Alberto and myself propose with consensus networks is just another point of view, but all of them are important and they need to be integrated.

Detecting Anomalies in Unmanned Vehicles Using the Mahalanobis Distance
Raz Lin

A model-free method based on statistical techniques to detect anomalies so that they can be corrected on time. They apply Mahalanobis distance (it sounds to me; i think that Alberto talked us about this for consensus networks; I have to check it). Using this distance eliminates the problems that appears in multidimensional data for euclidean distances. THe problem: it does not work well with qualitative changes (even if a partial order can be artificially defined?). An you’re losing a lot of domain dependent knowledge that can be useful!!

Dynamic ontology co-construction based on adaptive multi-agent technology
Sellami Zied

A tool to help ontologists to create ontologies (Dynamo project  A bit confusing: there is another dynamo for ontologies)

Tags: eumas09, agent, application, mwater, vehicles, ontology

Well, well, well, finally I’m in a room

Stigmergic Landmark Foraging
Nyree Lemmens, Karl Tuyls

Late

Integrating Organizational Control into Multi-Agent Learning
Chongjie Zhang, Shereif Abdallah, Victor Lesser

Problems of distributed learning

1. 1
2. 2
3. 3

Basic idea: organisational-based supervision framework. It’s a multilevel structure (recursive?) Lowest level network agents are ‘workers’. Each leaning agent reposrt its abstract state to its inmediate supervisor and them use rules a suggestions to transmit its supervisory informatio to its subordinates. Rules are set of forbidden actions and suggestions are actions with a degree in [-1,1], Rules are hard constraints and sugg are soft constraints that represent preferences.

The problem they’ve used to test this model is DTAP (distributed task allocation problem). Using a 27×27 agent grid… only!!!, too small!! I can manage several millions of agents to do the same :-( The results: interesting, but I don’t
understand all this stuff to be used in a small network as this: two supervision levels for such a group of agents.

It scales, but adding more supervision levels that may affect to the performance. I don’t like it. You’ll need a lot of layers for a really big network. Furthermore, in the experiments they’ve used a grid instead of a network and this is not ‘elegant’.

Multiagent Learning in Large Anonymous Games
Ian Kash, Eric Friedman, Joseph Halpern

We need to learn quickly, with minimal information and despite of noise. And to test their method they’re using games, but instead of being game theoretic games, they’re continuous, anonymous and designed games. He explains the method with a simple game but at the end it’s similar to game theory… I hate utility functions for agents. The behaviour can’t be reduced to a number or a function. Agents are more complex that that. We are more complex than that.

A simple algorithm to adapt the agent’s behaviour to the rest, so the dynamic converge despite of having agents making mistakes (so they’re introducing noise) in their decisions. As the number of agents increases, the system is more stable and converges faster… they’ve tried with 100 agents (again, too small for me). This results allows to tolerate strange behaviours.

Learning of Coordination
Francisco Melo, Manuela Veloso

Problem: many MAS solutions assume full joint sate observability because consider only local observability makes the problem too complex to be solved. But in many of these problems agent interactions are local. So they have to learn when interaction/coordination is advantageous. MDP and Q-learning is useed. And to show how it works, with an example of two robots that have to cross a gate.

They introduce a Coordination action (pseudo-action) and agents have to decide when to use this Coord action (it has a small penalty). Interesting method: agents can decide when to coordinate instead of exchange irrelevant messages all the time. They’ve tried with many different scenarios.

Abstraction Pathologies in Extensive Games
Kevin Waugh, Dave Schnizlein, Michael Bowling, Duane Szafron

Talking about poker competition for agents. Just two-playesrs. They use abstractions and the agent has to decice when to refine. Test with no-limit and leduc hold’em (small game, 6 cards deck, one ard ‘hidden’ and the other public). Boring… talking about the details of the game and many, many results.

State-Coupled Replicator Dynamics
Daniel Hennes, Karl Tuyls

Using evolutionary game theory, but it is single state dynamic, so it has to be extended to multi-state. Showing the behaviour in different classic games (as Prisoner’s Dilemma). Definitely…. i’m not interesting on this at all.

Wait a minute, with the examples I’ve seen hat it’s very similiar to our model of agreement, at least how it behaves. I’ll need to take a look to it. Too formal, but I hope that Alberto could help us with this.

Tags: aamas, learning, game theory

This is my first ‘normal session’ in the AAMAS. Yesterday we had a very boring poster session: three hours standing up just to talk 10 mins. with 4 people :-( And today an stressing morning with 6 parallel sessions Impossible to attend the all the topics I’m interested in.

On the Significance of Synchroneity in Emergent Systems
Adam Campbell, Annie Wu

I’ve arrive late, so I’m only trying to get a chair.

On Recursive Simulation
Latek Maciej, Rob Axtell, Bogumil Kaminski

Using something calles n-th order rationality (I haven’t the foggiest idea about what it is). Oh!, it’s about rationality in games (I don’t like game theory and any utility-based solution). They have a tactical model (that represents the evolution of the environment, without decision making information from other agents). So only the ‘trajectories’ of the policies can be found, nor the policies themselves. A n-th order rationality is defined recursively and… ups, too fast… he’s in the example now :-( playing Blotto (I don’t know this game). He’s talking about cognitive capabilities but, at the end, it’s represented just by one equation. Can knowledge be reduce just to one equation?

Adaptive Learning in Complex Evolving Trade Networks
Tomas Klos, Bart Nooteboom

Motivation: task allocation in networks of trading agents, with input/aoutput relations. IN classical economics, individual nodes are optimized, whereas transaction costr economics is focus on transactions (edges). Agents are rational and opportunistic. Buyers choose make or buy something and it’s implemented using Gale-Shapley algorithm for matching (game theory again… maybe the title of this session is wrong), using preferences based on scores related with potential profit and some trust (loyalty) measure.

Some experimental results, that turn into something interesting when they begin to consider the network itself, at least the indegree and outdegree (4 max. with a random network I guess). So, they have a model to simulate organizations using agents to check hypothesis.

A Mathematical Analysis of Collective Cognitive Convergence
Van Parunak

The idea of CCC is intgeresting: how a ‘closed’ collective can ‘corrupt’ the knowledge being something ‘endemic’. And they’re using agents-based models to explain why this is happening. They have simulated that and now they have a formal model of all this stuff.

An interesting result in theorem 4: it detects when the system converges (non deseable), so it can be corrected. And this convergence depends on number of agents, the number of topics and the topic’s density.

Emergent Service Provisioning and Demand Estimation through Self-Organizing Agent Communities
Mariusz Jacyno, Seth Bullock, Michael Luck, Terry Payne

A simulation model to match supply with demand in service based communities (coalitions and teams). Very difficult to synchronise choices in a centralised way and it leads to over-provisioning of services (againt a reason to create a distributed SF federation and, why not,  using a small world model ;-) This work is based on the emergent behaviour of insect colonies (OMG, ants again!) -> limited knowledge about peers and local behaviour (it sounds to me).

One interesting idea: to change the type of a provided service  has a cost, a penalty, so the system trends to keep the services unchanged as long as possible. Besides, resources are limited so a limited-size registry of known providers are maintained by customers, whereas providers have a limited-seize registry of user’s requests.

The simulation considers the amount of memory and only two types of services are available. When you have too few or too much memory the performance is worse that in a medium case, where you know enough providers/customers to work locally without service changes.

Effective Tag Mechanisms for Evolving Cooperation
Matthew Matlock, Sandip Sen

The last one, about considering expertise in a agent network using tag mechanism. Tags are a useful mechanism to promote collaborative behavior and it allows to reuse knowledge

Es el título de mi póster en el AAMAS ’09. Se trata de evaluar si es válido un modelo de red de tipo small world para distribuir a un grupo de plataformas  de agentes en una red de forma que se pueda localizar fácilmente dónde se encuentra un agente con el que nos queremos comunicar.

El modelo de red que se emplea ha sido propuesto por Kleinberg y garantiza que se pouede realizar un proceso de búsqueda voraz (tomando decisiones de forma local y sin volver atrás) acotado. A este tipo de redes se les llama redes navegables.

En un artículo más extenso lo he comparado con otros modelos de redes de tipo small world y la verdad es que sale bastante bien parado: es una red tan buena como la mejor (el modelo de Barabasi según mis pruebas) en cuanto a tolerancia a fallos (destrucción de enlaces en la red), pero mucho mejor en cuanto a la búsqueda, mejorando incluso a los modelos P2P.

Referencia

REBOLLO, M.: Small World Model for Agent Search (Short Paper).- In Proc. of 8th Int. Conf. on Autonomous Agents and Multiagent Systems (AAMAS 2009), Decker, Sichman, Sierra and Castelfranchi (eds.), May, 10–15, 2009, Budapest, Hungarytions.

Detrás de este título tan atractivo y motivador se enconde un post… acorde con el título :-) Básicamente la idea es tener algo que nos sirva como ejemplo para empezar a estudiar los espacios de acuerdos y su dinamica en un problema más “Consolider”. Y el ejemplo elegido es el modelado de una red fluvial. El propósito es poder usar los modelos de redes de consenso para tratar de modelar las interacciones en un sistema multiagente (SMA) para alcanzar un acuerdo. A través de los espacios de acuerdo puede llegarse a un conjunto de restricciones unidas a las leyes que regulan la propia dinámica del río. Y con algoritmos bien conocidos para resolver problemas de consenso en redes, a partir de su matriz Laplaciana

,

donde mantiene el grado de los nodos de la red y se define como

si y si

Para poder aplicarlo, es necesario disponer de un modelo de una red fluvial sobre el que se pueda aplicar este formalismo. El modelo de Scheidegger (1967) define la red como un grafo dirigido aleatorio sobre un retículo triangular. En cada intersección, se escoge al azar entre las dos posibles ramificaciones (izquierda o derecha). De esta manera, cada rama del rio es capaz de drenar una superficie de , donde es la distancia entre dos vecinos (es decir, que las distancias entre cada segmento de río es de -por simplicidad se suele escoger la unidad-).

Por otra parte, para determinar la relación entre el área de drenaje de un río y la longitud de su flujo principal se emplea la Ley de Hawk (1957):

donde l es la longitud del flujo principal, a es el área y el exponente de Hack h se calcula empíricamente y se encuentra en el rango 0.5-0.7. Con estos parámetros, es posible generar aleatoriamente redes que modelan un rio de forma realista, con lo que es posible generar distintos casos de prueba y comprobar empíricamente la validez de las propuestas.

Sobre estos modelos de rios, habrá que distribuir una red de agentes que representen las distintas entidades y personas responsables de la gestión y del consumo de los recursos hídricos. Creo que lo más adecuado es modelarlo como una red de tipo small world que simule las relaciones existentes entre los participantes. de esta forma es posible modelar las relaciones cercanas (entre regantes de una misma comarca) pero también la existencia de posibles relaciones lejanas que podrían modelar incluso a regantes de otras cuencas. Pero esta parte de la red social de riego la dejo para otra anotación

Para más información…

Dodds, Peter Sheridan: Geometry of river networks.- Thesis (Ph.D.)–Massachusetts Institute of Technology, Dept. of Mathematics, 1969. (PDF)

Referencias

J. T. Hack. Studies of longitudinal stream profiles in Virginia and Maryland. In U.S.Geol. Surv.Prof. Pap., 294-B:45–97, 1957.

A. E. Scheidegger. A stochastic model for drainage patterns into an intramontane trench. In Bull. Int. Assoc. Sci. Hydrol., 12(1):15–20, 1967.

Enforcing Security in the AgentScape Middleware
AgentScape is a middlewareto create agent-based applications and in this paper they’re showing how threads can be managed by using the security architecture for AgentScape. He begins with the architecture (see in AgentScape web wite).Two basic problemas: malicious hosts and malicious agents. SO basic sec. requirements are (i) authentification, (ii) integrity-migration- and (iii) authorisation.

Developing Intelligent Agents in Android Platform
Well, this is mine, so I’ll write it later :-)

Magentix: a Multiagent Platform Integrated in Linux
I know that, so I haven’t a lot of things to say. Magentix is an agent platform that are focus en scalability and efficiency, so it is quite integrated with the operative systems instead of using middleware (what will get him into trouble in a moment ;-). AMS (white pages) and DF (yellow pages) are distributed in different hosts and all that information is replicated (I hate that). Something interesting is how organisations are managed by the platform by the Organisational Unit Manager (OUM).

Only a question… what’s new on that? It sounds to me a lot!

From the assistants: about the RDF codification of messages (performance); problems: interoperability, integration with other tools, and AN ENVIRONMENT FOR PROGRAMMING DIFFERENT THAT VI!!!! ; other about scalability with AMS distribution in large systems (I said it!). And the last one, why comparing only with JADE.

Tags: eumas, agent, tools, agentspace., magentix, android, mda, mdd, pim, psm, jade-leap

An Agent Based System for Distributed Information Management: A Case Study
Sorry, I’m late :-(

Scholar Agent Alpha
Auction mechanism for authors (they pay for having their paper published). WHen they win the auction then they can use their “rights” to publish the paper. They are using a vickery auction (second price, closed envelope). the process is the following

1. call for papers
2. authors register themselves
3. authors bids (once)
4. system reports to winners and loser
5. wining paper is submitted for publication (and paid)

More cycles can be done, so agents can modify their bids and try to win.

Conservative Re-Use Ensuring Matches for Service Selection
Automatic retrieval, a semantic is needed (OWL-S, WSMO) and IOPEs has to be annotated. And flexible matches are very important (services that do not exactly match). But current semantic matchmaking considers services as a single operation. She proposes for the services to have roles inside a composed service. But for many of the operations are offered by patterns that are still unkown. And context usually is not taken into account, so the global goal can not be preserved, so they define an action-based representation of the operations of a service and an extension of re-use ensuring matches to produce substitutions that preserve goals. A modal logic is used; states are fluents and there’re 4 types of operations: one-way, notification, request-response and solicit-response. With those elements, choreography is defined through roles and all complex services will still works if conservative substitution is guaranteed. ANd this can be done if casual chains of actions are not broken when actions are sbustituted. What is important is that the problem of determining if a substitution is conservative is decidible.

A big “but”: youi need to have a choreography. Services can not discover service compositions, but working over a composite service and look for suitable substitutions.

Evolving Ontological Knowledge Bases through Agent Collaboration
Problem: an agent gives a queary about a concept c that it is not known, so the question can not be answered. Option 1: consult a mediator. Option 2: consult other agents. After that, some fragment of the ontology can be requested so the original agent can augment its knowledge and perform better interactions. But it has a cost and may not benefit the agent (it do not weigh the cost) ==> evolutionary method to augment ontologies with concepts related to its intereset domain. Two steps:

1. Merge Fragments: check which ones are related and merge them into one, create a powerset of all the axioms and, for each memmber of the powerset, check the consistence with existing ontology
2. Selection Process: find the average depth of the fragment, select the related properties level by level and incorporate them to the ontology.

This has been checked with an small number of agents (5 query and 10 speciliased) and several benchmarks. But htere’s no method for measuring the complexity of a T-Box.

Very interesting. One of the best papers I’ve seen these days.

Tags: eumas, agents, information, services, ontologies, owl-s, iope

Norms for Agent Service Controlling
About how add a normative system in an open system where agents and web services co-exist and interact. Agents are grouped in organisational units (OU), as coallitions or teams. OU provides services to the agents (or even the designers) to manage the organisation. They’re classified into 3 types: structrural (norm, role and unit creation), informational (about the current state) and dynamic (role adquisition).

Norms regulate the organisation and they’re divided into organisational and functional norms. She shows with an example the different type of norms, services and how all that works.

One thing that I don’t like is the definition of a new language. There’s well known logic specification and standards to formalise these concepts. A new language nowadays just complicate the interaction, even more for open systems…. and Pablo Noriega thinks as me :-)

Towards the Group FormationRecognition through Social Norms
Social norms are a good complement to laws, based on coordinated reaction on the members of the group. A norm is a set of observables and one action. How agent recognise as members of the same group? (that is, they have the same set of rules) They follow 4 algorithms: (i) basic (interact and save) (ii) the friends of my friends are also my friends (whitelist), (iii) blacklist (the opposite one) and (iv) labelling. Labels are put on the agents when they interact. They’re public: all agents can see them, but one agent can remove its own labels. False positives behave similar in the case of white and black lists. And it works better (obviously) for heterogeneous groups than for homogeneous groups. Very interesting. Take a closer look to that.

Integrating Image and Reputation Information in BDI Agents
3 approaches to control interactions among agents are possible: security, institutional and social (reputatin based). The last one can be modelled in a centralised way (as eBay) or decentrlised way (more interesting) and he’s going to present the Repage system to model image and reputation, integrated into a BDI agent. The difference between them is that image is my own view of the agent (experience-based), whereas reputation is what other agents say. A recommended lecture: the paper with six definitions for trust (look for it)

An Extendedc. Using Institutions To Foster Compliance in Open Multi-Agent Systems
Problem in open systems: heterogenous agents, with conflicting individual goals and limited trust. Some uncertainties (parametric -> environmental) can’t be reduced, but strategic ones can (decide if one agent has a bad behaviuour deliverately). In the case of anoinymous communities with esporadic interaction, it is very difficult to get efficient exchanges in the long run. But this can be solve even without superior detection capabilities by extending it. Well, she’s changing a bit the game because she adds a new player, but it is no serious.

Tags: eumas, agents, services, norms, organisation, institution

From Agents to Artifacts Back and Forth
In MAS, environment is considered as a monolithic and centralised component. This paper proposes a model based on artifacts. An artifact has (i) an user interface that provide operations for using it, (ii) the observable state, (iii) a mechanism to signal to external artifacts and (iv) a link interface to link to other artifacts.

But artifacts are not agents, they only react, they’re passive. And they are not objects either. A sample is shown in Cartago architcture. From the agent viewpoint, artifacts embedded operational and epistemic functions.

Towards Agent-Oriented Model Driven Engineering
To create MAS is a complex activity and many methodologies integrate MDE principles. MDE is based in the concept of model. They’re talking abut how differente methodologies, such as Ingenias or Adelfe are using MDE concepts. Some considerations: (i) concepts are still evolving, (ii) own semantic concepts, (iii) a wide range of levels of abstraction. They propose Model transformation By Example (MTBE), creating automatic transformation between pairs of models and shows some examples about how it works. The idea is to provide the tools with pairs of models that correspond to a transformation, so the systmen could learn from the and apply it to other cases. An very intersenting open issue-future work is how to include negative examples.

Reducing Communication Cost Via Overhearing
Showing how to reduce the communnoication cost by using mediators that filter invalid illocutions. The cost depends on the bandwith of the communication channel and the ammount of CPU (processing time) used to detect invalid illocutions. Interesting… and some ideas that I can use for SF distribution.

Towards Organizational Agent-Oriented Operating Systems
(Confirmado… Pixel sabe leer :-) Agent are complex systems but agents and even their platforms have no help from the operative system (OS), so they (we) are trying to define-and meybe to create-an OS that incorporate more complex abstractions to facilitate the execution of the agents.

Instead of being the process the basic exectional entity, agents organisations are included into the OS. Its based on a client-server view (I don’t like it), where there’re agents that provide services and others than ask from them. Services can be classified into two types: operations and resources. Operation serv. are sw services executed by an agent (open) and Resource serv. are hw services provided by devices.

What’s the difference between this proposal and middleware? I guess that it complicates a lot the OS design without giving clear advantages. I think that is better to provide basic management for agent concepts (as communication, roles, norms, organisations and so on), but we don’t need to have agents inside the operative systems.

Tags: agent, developtment, mdd, mda, communication, cartago, ingenias, adelfe

This evening begins with a panel session in which Marco Colombetti, Pavlos Moraitis, Pablo Noriega, Axel Polleres and Carles Sierra will talk about Agreement Technologies (AT).

AT is a paradigm for next-generation open distributen systems, based on the concept of agreement between computational agents. AT has a ‘cake’ (just as web services technology), that establishes the technologies needed for create such systems: semantics, norms, organisations, argumentation&negotiation and trust. All this activities are integrated in a COST action about Agreement Technologies (#IC0801). More information on the AT COST action web site. This initiative is very closely related with the Agreement Technologies group.

Semantics
Semantic aligment is an important issue in open environemnts, where agents meets and interact freeley. The main goals of this groups are

1. integration of ointologies with non-monotionic rules
2. quering over distributed ontologies
3. aligment with existing semantic web standards, enriching them with methods to enforce policies and trust online.

Norms
Reasoning about norms in MAS at design time and run-time to adopt, comply, enforce and modify normative systems. The three main challenges are

1. how to define formal, non-ambiguus and machine understandable nomrs
2. normative reasoning and negotiated  flexibility
3. usability of norms

Organisations
There’re meny interesting common points between organisations and agreements: an agreement is made in the context of an organisation, but meny times an organisation itself is the result of an agreement. Its main challenges are

1. how to model organisations
2. how agreements can be achieved in an organisational context
3. teamwork (team creation, build collective plans, corrdination or dissolution)
4. organisational change
5. implementation (architecture, patterns, environments&tools or verification)

Argumentation & Negotiation
There’s a lot of work done, but the interesting point is how aditional information is provided to the agents to try to convince them with adequate arguments. This way can facilitate the achievement of agreements. Pending issues are

1. the notion of concession is not modelled
2. how agents choose the and offer to propose in each dialogue step
3. deeper investigatiom on the notion of tactics and strategies
4. decision-making models to reach satisfactory agreements

Trust
It is very important after the agreement has been achieved, when agents decide to honour the agreement or not. The idea is focus on groups working on semantics, norms and social models to model trust. Important things to think about are

1. scalability (social network models, large scale systems, negotiation)
2. semantics (not defining a single global ontology, but deal with misbehaviours and misundestandings)
3. similarity (how to use past experiencies to assest how much ‘trust’ you can put non an agreement)
4. balence between norms and trust (the more norms you have, the less risk the opponent has and the less trust measures has to be taken)

After the exposition, some conclusions and questions.

Tags: agreement, agents, semantic, norms, organisation, argumentation, negotiation, trust