Miguel Rebollo

A Trust Aggregation Engine that Uses Contextual Information
Joana Urbano

Enhacing traditional aggregation approaches by the inclusion of dynamics and being situation-aware (context).

Dynamics, based on Hysteresis of Trust and Betrayal (Straker, 2008), introducing 3 properties as asymmetry, maturity and distinguishable past. Thier model is called SinAlpha. Asymmetry penalizes intermittent behav., maturity avoids selection with few evidences and distiguishably prevents fast forgiveness. But the sin approach is not better than the linear model (experimentally).

Situation-aware is covered by contextual fitness. It is something very similar to CBR: clustering, stereotype extraction, analysis of similarity. Interestng for me: Multidimensional context representation for situational trust (Rehak, Gregor & Pechoucek, 2006). To do: deal with newcomers (first-encounter)

Preliminary Results on Reputation Systems: Balancing Quantity and Quality
Jonathan Ben-Naim

Agents in a network with a ranking that models reputation. A global measure (untractable for very large open systems as web). Two axioms: transitivity and strict transitiviy (good as first approach, but it can not be generalized). They refines the values in different interactions. What does it happend when there are loops? A lot of things to explain: ‘random’ initial ranking that he promises is not affecting to the final result (I can’t believe this), no weight/importance, the use of group size (stricti trans.) is questionable….

An Interaction-oriented Model of Trust Alignment
Andrew Koster

Well, I’ve seen this a lot of times: how to align trust concepts. Particularly, I prefer to have an standard on this part. Because we can continue with that: for example, ACL; why can’t we align ACL ontologies so the agents can speack in any language? In this case: agents share the same sintactis (about trust) and the semantics has to be aligned (what does it means to have a 0.8 confidence?) Implemented using inductive llogic programming (scalability?)

Supplier performance in a Digital Ecosystem
Angela Fabregues

Deals with partnership selection in cases where negotiation/argumentation is involved.

(Inciso: OMG bolitas paseándose por la pantalla, a DocThreeC le encantará esta plantilla ¿se lo digo? luego nos torturará).

To define the model of trust she begins with the ontology, similarity, expectations (see the invited talk of Carles Sierra this morning). By using past experiences, the probab. distribution of expected observation is modified. So, at the end, you do not take into accout the information about the exact object, but the similar ones too.

(otro inciso: ¡qué garrillas tiene la chica de las traspas! parece Ana Obregón :-D)

She continues explainig how the trust value is calculated using all these things and a bit (quickly) about similarity.

On Norm Internalization. A Position Paper
Daniel Villatoro (as guess star) in behalf of Rosaria Conte

How agents internalize existing norms and incorporates them to their behavior. He begins talking about goal internalization. At the beginnig, you behav. is directed by norms, but when yo asimilated them then you behaves in that way not to avoid a punishment, but because you want to behave in that way (for example, to stop when light is red in a semaphore). Too fast to listen and to write at the same time (you know, I’m a man), but it is a very interesting thing and I think that is related with adaptive organizations. I have to read the paper this this idea on mind. An intertesting point: urgency is a factor that affects to the speed at which intentions are internalized.

It is integrated with EMIL-A (BDI), N-Bel -> N-Goal -> N-Intentions that are Internalized as a conformed behav. A comment: this work is about people, not artificial agents.

Tags: eumas09, trust, reputation, dynamics, context,

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

Invited talk by Carles Sierra

The overview: beginning from individual opinion, how it is transformed into group opinion to achieve some reputation. Reputation is actually a grupal opinion about someone or something.

The example: Liquid publishing, about how people change their opinion by arguieng and considering reputation (the example is about paper review for a conference). The reputation of author, reviewers, the paper or the conference itself are considered.

First step: forming individual opinions. Agent receives messages that contribute to agent’s knowledge creating a distribution about the quality (or the true value) of the predicate. And this quality decays with time, having a decay limit distribution. This process (updating information and decaying with time) is reactive reasoning (giving formula about all these things). But we can have two types of opinion: verifiable (tomorrow will rain) and unverifiable (Earth will exist in 1 million years).

Second step: To structure the knowledge (as we organize sections in a paper). So the opinions can be given about each one of the identified elements and these opinions can be used to create a reputation value. So opinions are associated to nodes in the structure.

With that, entering in the third step: how group opinion are formed. To do that, (i) a language to share opinions, (ii) distances about opinions and (iii) methods to aggregate opinions are needed. So we are looking for a function that summarizes the group opinion. About the language, we can inform opinions (somehow a subjective thing) and experiences (objective facts). About getting information, citations can be good, but opinions can be even better (problem with ‘the rich get richer), and this can be done using distances between distributions: (i) by calculating how similar two functions are, (ii) calculating the distances between them (EMD)-not euclidean for opinions- Explaining different methods to combine opinions. Something interesting: reputation labels; inexorable, predetermination, persuasiveness, compliance, dogmatic and adherence that describe the position of an agent with respect to the opinion ofa group. BUt individuals have social relationships and this information ca be also taken into account.

So, to summarize:

• reputation is becoming crucial for all sort of web-related applications
• current model ignore the structure of the knowledge, as social relations
• integrated models that deals with all these information (social, structural, dependence…) are required

Tags: eumas09, reputation, opinion, group, knowledge

d

Semantic Web Service Composition using Planning and Ontology Concept Relevance
Ourania Hatzi

A planning approach to compose services… very similar to our work: OWL-S to describe services, Pellet as semantic reasoner for OWL and a planner to compose services. The most interesting thing is not about services, but about semantic awareness when the planner fails.

Improving Web services adaptability thanks to a synergy between aspect programming and a multi-agent middleware
Flavien Balbo and Valérie Monfort

About enterprise IS. They add “aspects” to ws and use a middleware (agent-based) to improve adaptability. I don’t understand why all these mess: AOP to add dynamically new behavior to web services and agents just to coordinate (by communication) ws… I guess that using just one technology will be easier.

QoSS Policies Operating for Web Services within SOA
David Allison

QoSS: Quality of Security Service. Metadata includes information about authentication, authorization and privacy, embedded inside the SOAP message. I don’t know why to modify SOAP is needed instead of use the WS-Auth, WS-Privacy and so on, most of them provided by the service platforms. This can be useful for “private” service (inside an intranet, or for enterprise services), but not for public ones.

Building Blocks: Layered Components Approach for Accumulating High-Demand Web Services
Satoshi Morimoto, Satoshi Sakai, Masaki Gotou, Heeryon Cho, Toru Ishida, and Yohei Murakami

They use web services as language resources that can be combined to create new tools (A Language Grid). For example, multiple dictionaries. Building blocks for simple components programed in PHP or Java are encapsulated inside web services. This is used to create new systems (I assume that it’s used by programmers) faster than without them.

A framework to guarantee time-bounded composed services
Elena del Val.

Well, these is our paper, so I have no things to say…. it’s perfect :-D Ok, ok… I tell you something about that. The idea is to guarantee service execution time (soft real-time). Commitment Manager reach agreements with providers to get a temporal windows in which is ensured that the service will be provided. WS-Agreement protocol is used to do that. 

Supporting Web Service Protocol Changes by Propagation
Ahmed Azough

About business protocols. They consider them as FSM and allow add and delete state, transitions and the final state.

Reasoning about Web Services with Local Closed World Assumption
Limin Chen, Hong Hu, and Zhongzhi Shi

Ups, no one for this paper, so we’ve finished

by Ronald Yager

Fuzzy sets to represent linguistic concepts, reducing the human.machine gap. A common vocablary is needed, so human can use linguistic terms and machines can use fuzzy representation to represent the same concepts and understand each other. For example: about Age {young, old, senior, 23, about 40}

Granular computing is a technique that can be used to link linguistic (human) and mathmatic (machine) concepts. It extensd the capabilities for analyzing social relational networks by enabling tue use of human like concepts with fuzzy sets and granulat technologies. A Social network is actually a set object…. a lot of maths and concepts about the structure of the graphs in social networks now… I prefer to listen instead of to write, sorry.

Invited talk by Marco Dorigo.

Swarm intelligence: IA technique based on emergence of grupal behaviors from a decentralized, individual intelligences. Examples from biological societies: ants searching paths, transporting things and assembling things to build a bridge. And they’re many other examples about foraging or division of labor.

Swarm robotics motivation: fault tolerant system, high parallelism degree, scalability and low cost. A swarm-bot is a robotic system composed by a set of robots that can touch each other and create more complex structures. Explainig hardware stuff now. And some videos seeing them at work.

Basic behaviour can be hand-coded or it can be evolved from simple neural networks and it can be loaded in areal s-bot. Capacities:

• coordinate motion: detect the movement of the rest and determines its behavior. Combined with self-assembly allow the robots to create special structures to avoid obstacles
• selft assembly 
• cooperative transport: move things bigger than the robots. Again, can be combined with self assembly. OMFG!! They’re moving a child!!!
• goal search and path formation: limited sensing acapabilities, so they moves randomly until they sense the goal they begin to form a chain. Some kind of bread crumbs :-)

Ongoing work: functional self-assembly, morphology formation and swarm-level fault detection. Fantastic

A session with short papers, so I have to be ready.
 
Methodologies for self-organising systems: a SPEM approach
Mariachiara Puviani, Giovanna Di Marzo Serugendo, Regina Frei and Giacomo Cabri
 
Late :-( I,m sorry But it seems interesting. Take a look at the paper
 
Self-organization of peers in agent societies
Martin Purvis
 
Goal: to investigate gossip-based mechanism for self-organization of agents divided into groups: Decentralized, scalable and with partial information. The problem domain used is sharing digital goods in electronic societies. Can it work in a decentralized way (without supervisor)?
 
Sharing incurs a cost: the donor have a cost and the receiver receives the benefit. How agents can be “forced” to cooperate and share goods? First 100 iterations, agents play and gossip. After 100 iterations, they can move to other group.
 
Showing results about how groups are finally separated and well-defined. Why are agents leaving a group? (i) tolerance level is met (others do not share enough) or (ii) benefits are not improving.
 
 
An Autonomy-Oriented Paradigm for Self-Organized Computing
Jiming Liu
 
Typical application areas of AOC: hard problems and complex systems. Related work: discrete-time propagation models or immunization strategies for restraining virus spreading (hey, guy, it’s a short paper… when are you going to begin?) OK. decentralized search for immunization problem. The entity tries o find the node with the highest degree in its local environment. If these nodes are protected, the virus spreading can be delayed or even stopped. Possitive feedback is used to rapidly discover a good set of high-connected nodes at early stages.
 
 
Simulation of the Rungis Wholesale Market: lessons on the calibration, validation and usage of a Cognitive Agent-based Simulation
Philippe Caillou, Corentin Curchod, and Tiago Baptista
 
Justa paper about how a concrete problem has been solved (simulated) using agents. Not interesting for me. Sorry.
 
 
Silicon Coppélia: Integrating three affect-related models for establishing richer agent interaction
Matthijs Pontier and Ghazanfar Siddiqui

Goal: a robot that could interact with humans (I guess, Because I hardly can hear him… a microphone! good). Emotional models formed by the integration of 3 models: CoMERG, I-PEFiC and EMA. It uses a emotion regulation model (Gross). These approaches models important aspects of human affective behavior, but all of them miss out something important.

Proposal and simulation experiments. Something that wonders me is how utility values are calculated and emotions can be perfectly described just by a number. That’s because I don’t believe in utility-base models (I can model this as “utili-base models likelihood = -1″ :-) Anyway, examples seems very complex and with a sufficient number of variables/parameters.

So they can model things as irrational decision or emotions based on believes quite well. Interesting.
 
 
Transition Process Distinction in Multiagent Organization
Eric Matson

A low-level paper… (hardware related, I mean :-) How organizations can be embedded on physical devices? Over the time, the organization evolves from its initial state untilo it reaches some state of global satisfaction.

All is based on the concept of transition rather than reorganization. Computationally, they ae very different problems. I must read this paper. It sounds interesting and promising. Furthermore, can be useful for our Android agents.
 

CCR : A Model for Sharing Reputation Knowledge Across Virtual Communities
Nurit Gal-Oz

Working with private identities of separate virtual communities (Tric -Deustche Telecom-) and reputation mechanisms to share information among different communities. The process is formed by 3 blocks
1.- Enabling preconditions

• first
• category matching level: [0,1] value representing community correlation based
• third
• domain confidence

Interesting: one of the measures ids is based on Shannon entropy of the domains (I need o revise that again)

2.- Conversion of reputation values
3.- Attribute matching (some kind of ontology alignment) as a [0,1] value, with some confidence level (certainty)
Showing an example of travel agencies :-), looking for a hotel in Milano, using communities in Trip Advisor, Expedia and Booking.

Monetizing User Activity on Social Networks – Challenges and Experiences
Meenakshi Nagarajan, Kamal Baid, Amit Sheth, and Shaojun Wang

Well-known monetization models for the Web, but not easy on Web 2.0 and SNS. Actually, advertisement-based models are used with marginal success, due to (i) informal nature of content, non-policed content and because people are there to network. So, at least, the ads have to consider: (i) identify monetizable posts (intents behind users posts) and (ii) identifying keywords on user’s comments.

Consider that people write sentences, not keywords or phrases, so the system has to be ready to analyze that to locate action patterns around entities. The example: how people seek for information. Patterns used are “where do I find…. does anyone know how… someone tell me where…” So all these patters matches with a seeking pattern in the Candidate Pool and the system can identify the question (this is very similiar to dialog characterization for agents -see AIWS slides-)

Monetization potential is calculated from this seeking score and calculating a transactional intention score for each one of them.  All test has been done off-line!! What about the cost? Because it’s important to do that on-line.

A Composite calculation for author activity in Wikis: accuracy needed
Janette Lehmann

Interesting title, but I barely can hear her :-(… oh!, better with the microphone.

Analysis of social interaction spaces (wikis, blogs, twitter…) evaluation activity, dynamics, identifying communities and topics, so they can improve existing infrastructures. They use SONIVIS as visualization tools. And today she is going to speak about wikis. Motivation: to evaluate author activity. They activity is characterized by (i) the number of changes/versions, (ii) a betweenness centrality measure among authors, (iii) significant content: frequency an author has added a term to a page and the importance of this term. This measure is normalized. These values are combined in the final author contribution. This measure is dynamic, so a cumulative author contribution can be calculated.

The example uses the English Wikipedia articles about virtual reality. A six-month period has been analyzed. Results are in the paper.

Model for Voter Scoring and Best Answer Selection in Community Q&A Services
Chung Tong Lee

The problem: how to select the best answer in Q&A communities (as Yahoo!Answers). A Voting Score mechanism is presented, based on a fixed point basis.

Voting is affected by (i) social bias -vote based on answerer, not in the quality of the answer, (ii) personal gain. The formula for the voting and the fixed point characterization is presented (see the paper). Some examples to model the voting (simulations) using random voting and ballot stuffing simulation. Zipf’s law used for generate random votes  (entropy measure). Results again in the paper.

invited talk by Michael N. Huhns

Interesting metaphores for differente technologies

• DPS: decomposition
• DAI: coordination
• MAS: interaction
• SOC: encapsulated functionality with a public interface

Nowadays, some social challenges in economy, energy, environment, transportation, telecommunication are the great problems of our age. And they’re massive, distributed, many-faced, with a large number of dependant componentes, controlled action is needed, but centralised control infeasible. SO agents are the tool (I guess) for addressing these problems.

Characteristics of agent paradigm:

• large-scale multiagent participation
• spatially distributed
• temporally extended
• uneven progress
• possibly cooperative
• design domain isomorphic to execution domain
• constrained: it can’t pave everything and no semantic mismatches
• solution is not centralised, bat it occurs at the edge

Example: individualised transportation.
routes of rails and traffic are designed centralised by engineers, instead of be done in real.time by passengers. Speeds limits are set centrally andd fixed. Traffic lights are barely reactive to local traffic, when it can be auctioned in each intersection…. and meny other examples.

Example: individualised healthcare
The systems are designed for hospitals and caregivers, but not for patientes

Example: grocewry shopper
supermarket chains use IT to set prices, but they’re no systems for shopper to find fair prices. Even shoppers could use RFID tagged items in their own profit.

Example: governance
a citizen has a vote that is given to a representative to be used for N issues.

Example: energy
Europen S-TEN project is using semantic web tech to make each componente of hte energy grid to report on its status intelligently. The result is a finer grained status to human operators of the grid

Example: taxation
determine the fair share of every one. In general, people doesn’t mind to pay what is fair, but it is diffucult to determine in the case of ‘commons’.

…and a very interesting example in logistics that I’ve prefered to listen to).

All this is compared with the example of Columbia university: they put the building rounded completely by grass. And after one year they just paved the worn paths made by people. This is the same criteria in all the previous examples: let the agents to interact and to create or re-create the model by themselves.

Consensus …
Consensus ontology: a first step towards agreement spaces. Take a look at this.
Consensus behaviour: select a plan/sequence of actions from the behaviour of the rest (emergence?). So you can find the best algorithm to do something

Hyperscale sw development: consensus provides a different way for developing sw: encourage lots of people to contribute to software systems and they use all of their constributions. The problem is how all these contributions can be combined.

Idealised SOC: given the requirements of an applitacion: (i find a sert of services that cover the requirements nad (ii) workflow
but it is still unused. They’re very similar to agents, but agents have some benefits: autonomy, they’re active components, are complex (n-party interactions)…The problem: flexibility and reliability. HOw to fix it?
decrease autonomy increases predictibility n(this is why SOA is more used that agents)
removing semantic inconsistences too
transaction concepts can ensure ACID resutls
agents can recover states and maintain their progress toward overall goals at run-time.

Well, and it ends with a great sentence that I’ll add here. I’ve written this post complelety on-line, so I’ll need to review it and make some (minor) changes.

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