************************************************************************** NEWSLETTER ON DECISION AND REASONING UNDER UNCERTAINTY Issue 98004 Editors: Salem Benferhat, Henri Prade 25.11.1998 Back issues available at http://www.ida.liu.se/ext/etai/dru/binf.html ************************************************************************** This newsletter contains: 1. Call for contributions 2. Discussions regarding the paper of D. Poole: a contribution from H. Geffner 3. A reply from D. Poole to the comments of H. Geffner 4. Continuing discussions: A reply from H. Geffner 5. An information from J. Halpern Annoucing A Computing Research Repository 6. New books 7. Table of Contents for Applied Intelligence Vol. 9, Number 2: Special Issue on Systems for Uncertain reasoning 8. Conference announcements ========================================================================== 1. Call for contributions ========================================================================== Thank you for subscribing. We are happy to report that about 400 people have now personnally registered for receiving this Newsletter. We encourage you to take advantage of this rather large special interest and discussion group and to submit papers to ETAI-DRU Journal. ETAI-journals aims at receiving high quality works. Note that the papers remain accessible (with the contributions of their discussants) for 25 years (even for papers which are finally not accepted in the ETAI Journal). We recall that the call for papers is available at : http://www.ida.liu.se/ext/etai/dru/index.html Besides, please send emails to benferhat@irit or prade@irit.fr for announcing Conferences, Books, Journal issues, PhD thesis and Technical Reports, Career Opportunities and Training, and Softwares dealing with uncertainty. Books reviews are also welcome. Feel free to contact us for any further information. ========================================================================== 2. Discussions regarding the paper by D. Poole: a contribution from H. Geffner ========================================================================== David, I've enjoyed reading your paper "Decision Theory, the Situation Calculus and Conditional Plans". I find it hard to disagree on your general point that a general model of action must combine elements from logic and probability/utility theory, so my questions and comments are addressed more to the question of how this combination is achieved in your proposal. I'll try to comment first on the more general issues. 1. In page 24, you say "The representation in this paper can be seen as a representation for POMDPs". If so, I wonder, why not present the representation in that way from the very beginning? Wouldn't that make things much clearer? Namely, a POMDP is described by (see [1]): A - state space B - actions C - transition probabilities D - cost function E - observations F - observation probabilities Then the constructions in your language could be seen as a convenient/concise/modular way for specifying these components. Actually, once the mapping from the language to these components is determined, no further semantics would be needed (i.e., the probabilitities of trajectories, the expected utility of policies, etc., are all defined as usual in POMDPs). This is actually the way we do things in [2]. 2. I can see some reasons why not to present your representation as a "front end" to POMDPs. - you want to be more general; e.g., be able to deal with cost functions that depend on the system history. If this is the case, I'd suggest to introduce "generalized" POMDPS where (cumulative) cost functions do not have to be additive (btw, in such POMDPS belief states are not necessarily sufficient statistics ..) - you want to accommodate multiple agents. Yet this is not done in this paper, but even in that case, multi-agent POMDPs could be defined as well, and have been defined in the completely observable setting (e.g., see Littman). - you are not interested in determining an optimal or near-optimal solution of the POMDP but are interested in analyzing the expected cost of a policy supplied by a user in the form of a contingent plan. Again, this is no problem from the point of view of POMDPs, as long as the contingent plan determines a (stationary or non-stationary) *function* of belief states into actions (see [4]). Indeed, you require *more* than this when you demand (page 18), that the tests in conditions of the plan, be *directly observable* at the time when the conditions need to be evaluted. This is not required in POMDPs [4] or in possible world accounts [3], where the test may be known indirectly through other observations. 3. A final comment about a non-trivial problem that results from the decision of removing all probabilities from state transitions transferring them into suitable priors in "choice space". Consider the effect of an action "turn_off" on a single boolean fluent "on" with transition probabilities: P( on=false | on=true; a=turn_off) = .5 P( on=false | on=false; a=turn_off) = 1 Let's say that initially (on = true) and then we perform a sequence of N consecutive "turn_off"s. Then the Probability of (on = false) is .5^N, which clearly depends on N. I don't see how this behavior could be captured by priors over choice space. This seems to be a strong limitation. It looks as if probabilistic actions cannot be handled after all. Is this right? Refs [1] L. Kaebling, M. Littman, and T. Cassandra. Planning and Acting in Partially Observable Stochastic Domains, AIJ 1998 [2] H. Geffner, B. Bonet. High level planning with POMDPs. Proc. 1998 AAAI Fall Symp on Cognitive Robotics (www.ldc.usb.ve/~hector) [3] H. Levesque. What's planning in the presence of sensing? AAAI96 [4] H. Geffner and J. Wainer. A model of action, knowledge and control, Proc. ECAI 98 (www.ldc.usb.ve/~hector) ========================================================================== 3. A reply from D. Poole to the comments of H. Geffner ========================================================================== > David, > > I've enjoyed reading your paper "Decision Theory, the Situation > Calculus and Conditional Plans". I find it hard to disagree on > your general point that a general model of action must combine > elements from logic and probability/utility theory, so my questions > and comments are addressed more to the question of how this > combination is achieved in your proposal. I'll try to comment > first on the more general issues. > > 1. In page 24, you say "The representation in this paper can be seen > as a representation for POMDPs". > > If so, I wonder, why not present the representation in that way > from the very beginning? Wouldn't that make things much clearer? One of the problems with writing a paper that is trying to bridge different areas is to try to explain it for all readers (or to keep them all equally unhappy). I suppose I have succeeded when POMDP researchers say "this is just a representation for POMDPs" and when situation calculus researchers claim "this is just the situation calculus with some some probability added" or those studying Bayesian networks claim that "this is just a rule-based representation for dynamic Bayesian networks with actions and observables". Describing it explicitly in terms of POMDPs may have made it easier for POMDP researchers, but not necessarily for everyone else. > Namely, a POMDP is described by (see [1]): > > A - state space > B - actions > C - transition probabilities > D - cost function > E - observations > F - observation probabilities > > Then the constructions in your language could be seen as a > convenient/concise/modular way for specifying these components. I thought I did do this. (At least I was trying to do this, and I knew when I was finished when I had defined all of the components of a POMDP). The actions and observables of Definition 2.6 are the same as B and E. I specify C, D & F using rules. In particular I represent the cost function in terms of rules that imply the $utility$ predicate (Section 2.6), and observation probabilities in terms of rules that imply the $sense$ predicate (Section 2.7). > Actually, once the mapping from the language to these components > is determined, no further semantics would be needed (i.e., the > probabilities of trajectories, the expected utility of policies, > etc., are all defined as usual in POMDPs). > This is actually the way we do things in [2]. That is essentially what I did. There are some problems with just relying on the definition of a POMDP. This has to do with what is a state. In the formal work, the notion of state is taken as primitive. However, for a well-defined semantics we need to be clearer. A state could either mean: (a) everything that is true at a stage. This would include the actual time (if it is relevant), the values of sensors, the accumulated reward, etc. (b) a sufficient statistic about the history to render the future independent of the past given the state (which, by definition is Markovian). This typically doesn't include the time, the output of the sensors, or the accumulated reward. I have taken the approach of (a). Most POMDP researchers take the approach of (b) because their algorithms reason in terms of the state space. They therefore want to keep the state space as small as possible. I don't want to reason at the level of the state space, but instead at the level of the propositions. > 2. I can see some reasons why not to present your representation > as a "front end" to POMDPs. In general I don't want to do this, because I don't think that reasoning in terms of the state space is the right thing to do. What we can learn from AI is that we want to reason in terms of propositions or random variables, not in terms of the state space. One of the themes of my work is to find compact representations and to exploit the compactness for computational gains. Thus, I don't want to present this as a "front end" to POMDPs. If you mean following the semantic intuitions of POMDPS, then I think I do. But POMDPs are just one of the formalisms I am building on. They have nothing to say about useful independence assumptions or about concise descriptions of actions (state transition functions), which is what this paper is mainly about. > - you want to be more general; e.g., be able to deal with cost > functions that depend on the system history. If this is the case, > I'd suggest to introduce "generalized" POMDPS where (cumulative) > cost functions do not have to be additive (btw, in such POMDPS > belief states are not necessarily sufficient statistics ..) (unless you include the accumulated reward in the state.) > - you want to accommodate multiple agents. Yet this is not done > in this paper, but even in that case, multi-agent POMDPs > could be defined as well, and have been defined in the completely > observable setting (e.g., see Littman). This was done in my AIJ paper, "The Independent Choice Logic for Modelling Multiple Agents Under Uncertainty". I actually don't think that the ICL_SC formulation is as good for this as the formalism in that paper. > - you are not interested in determining an optimal or near-optimal > solution of the POMDP but are interested in analyzing the expected > cost of a policy supplied by a user in the form of a contingent > plan. Again, this is no problem from the point of view of POMDPs, > as long as the contingent plan determines a (stationary or > non-stationary) *function* of belief states into actions (see [4]). First let me point out that POMDPs are not defined in terms of belief states. It is a theorem about POMDPs (Astrom 1965) that says that an optimal policy can be represented in terms of a function from belief states (probability distributions over ordinary states) into actions. However, "optimal" here doesn't take into account computation time. It is pretty obvious that this isn't true for bounded optimal agents (where the computation time and the space are taken into account), which is where I would like to take this. I *am* interested in determining an optimal or near-optimal solution (but one that takes computation time and space onto account). I want to be able to define the expected utility of any strategy the agent may be carrying out, whether it maintains a belief state or not. No matter what program the agent is following, it can be described in terms of a conditional plan that describes what the agent will do based on the alternate sequences of observations (i.e., its function from observation sequences into actions). Note that this doesn't refer to anything inside the agent. The agent could be reasoning with a probability distribution over states, remembering a few previous observations, just reacting to its current observations, or actually following a conditional plan. The important thing is what it does is based on what it observes. What I am advocating is good hacks; the best agent won't do exact probabilistic reasoning. We need good strategies, perhaps like your RTDP-BEL. However, it is not obvious (to me anyway) that the bounded optimal solution will necessarily be an approximation to the (unbounded) optimal solution. The best real (bounded time and bounded space) agent may do something quite different than approximating belief states. Unless we have some representation that lets us model the expected utility of an agent following its function from observation and action history into actions (i.e., its transduction), we won't be able to specify when one agent is better than another. The conditional plans (policy trees) let us model this, without any commitment to the internal representation of the agent. > Indeed, you require *more* than this when you demand (page 18), > that the tests in conditions of the plan, be *directly observable* > at the time when the conditions need to be evaluated. > This is not required in POMDPs [4] or in possible world accounts [3], > where the test may be known indirectly through other observations. NO! I mean that the conditions are those things that are observable. I mean whatever evidence (the "other observations") that the POMDPs or possible worlds accounts take into account. This is a standard technique in Bayesian networks/influence diagrams; we create a variable that represents the actual value sensed. I am quite happy to have this as a (state) variable. You don't want it as a part of your state because it isn't necessary to make the system Markovian. This is fine, but I can model these "other observations" using the "sense" predicate. Is the sense predicate part of the state? I don't know; that is one reason why I didn't blindly accept the notion of state. > 3. A final comment about a non-trivial problem that results from > the decision of removing all probabilities from state transitions > transferring them into suitable priors in "choice space". > > Consider the effect of an action "turn_off" on a single boolean > fluent "on" with transition probabilities: > > P( on=false | on=true; a=turn_off) = .5 > P( on=false | on=false; a=turn_off) = 1 > > > Let's say that initially (on = true) and then we perform a > sequence of N consecutive "turn_off"s. Then the Probability > of (on = false) is .5^N, which clearly depends on N. > > I don't see how this behavior could be captured by priors > over choice space. This seems to be a strong limitation. > It looks as if probabilistic actions cannot be handled > after all. Is this right? Here is how I would represent this: Facts: on(do(turn_off,S)) <- on(S) & off_failed(S). on(init). C_0 contains {off_failed(S),off_succeeded(S)} for all situations S P_0(off_failed(S)) = 0.5 for all situations S [Note that the completion of the facts for on gives your two conditional probabilities.] We can then derive P(on(do(turn_off,do(turn_off,do(turn_off,init)))))= 0.5^3 just as you want (similarly for any N). We need an assumption for each time step. The minimal explanation for on(do(turn_off,do(turn_off,do(turn_off,init)))) is {off_failed(init), off_failed(do(turn_off,init)), off_failed(do(turn_off,do(turn_off,init))}. In all worlds with this explanation true, the light is on in the state after three turn_offs. The probability of these worlds sum to 0.5^3. Any probabilistic actions that you can represent in a dynamic Bayesian network, I can represent in the ICL_SC. Moreover I can represent it at least as succinctly as the dynamic Bayesian network (up to a constant factor) and sometimes exponentially (in the number of state variables) more succinctly (when there is context-specific independence). The mapping is local and modular. Thanks for taking the time to read and comment on the paper. I hope my comments help makes the paper and what I am trying to do clearer. Let me stress that this paper has nothing to say about computation. Whereas your work was motivated by being a front end to an actual POMDP algorithm; the motivation here relies on being able to deliver on POMDP algorithms that can exploit the conditional and contextual independencies of the representation. Unfortunately, I can't show you such algorithms now. But we are working on it. David ========================================================================== 4. Continuing discussions: A reply from H. Geffner ========================================================================== David, thanks for your answers; they helped me a lot. Just a brief follow up on the status of your proposed framework: a new model? a new language? a new algorithm? all of them? ... Best regards. -hector > > 1. In page 24, you say "The representation in this paper can be >> seen as a representation for POMDPs". > > > > If so, I wonder, why not present the representation in that way > > from the very beginning? Wouldn't that make things much clearer > >One of the problems with writing a paper that is trying to bridge >different areas is to try to explain it for all readers (or to keep >them all equally unhappy). I suppose I have succeeded when POMDP >researchers say "this is just a representation for POMDPs" and when >situation calculus researchers claim "this is just the situation >calculus with some some probability added" or those studying >Bayesian networks claim that "this is just a rule-based >representation for dynamic Bayesian networks with actions and >observables". Describing it explicitly in terms of POMDPs may > have made it easier for POMDP researchers, but not necessarily > for everyone else. I think I don't agree with this. I believe that POMDPs are a general and natural *model* for sequential decision problems that involve sensing. They are not for POMDPs researchers only; in the same sense, that logic is not only for logicians. And they are simple too (unlike some of the POMDP *algorithms* that are indeed complex). I think that anyone dealing with *sequential decision problems that involve sensing* should know about POMDPs, whether or not they appeal to POMDP algorithms for solving them, and whether or not they deal with probabilities. I know this sounds dogmatic ... but it is the truth!!! :-) No, really, POMDPs are very useful for understanding and identifying the different dimensions of a decision problem: transitions, costs, information; and at the same time they have little to do with probabilities (namely, even if all probabilities become zero or one, POMDPs are still very meaningful and don't reduce to anything else as 0-1 probabilities that would reduce for instance to logic). On a more general note, I think there are three essential aspects to the work in planning and control in AI: - models - languages - algorithms Models are about the mathematics of the task: what is a problem? what is a solution? what is an optimal solution? Languages are for describing these models in a convenient way. Algorithms are for computing the solutions. I think these three ingredients are always present in approaches to planning and control in AI, and I believe it is useful to make to them explicit; even when the algorithms may take advantage of the particular language in which the model is represented (as in Strips planning). I understand that you expect your language to be useful not only for specifying POMDPs in a convenient way, but also for solving them conveniently. That would be great. Yet even in that case, people using different POMDP algorithms could in principle benefit from your language for setting up their POMDP models. This could also be important, and indeed, in the recent AAAI Symp on POMDPs, the need for good languages for building POMDPs for particular applications and for exchanging benchmarks was emphasized. May be your language, as well as other action languages suitably extended, could fill up that need. As you know, we have also been doing work in that direction. Best regards. - hector ========================================================================== 5. Annoucing A Computing Research Repository ========================================================================== Researchers have made their papers available by putting them on personal web pages, departmental pages, and on various ad hoc sites known only to cognoscenti. Until now, there has not been a single repository to which researchers from the whole field of computing can submit reports. This is about to change. Through a partnership of ACM, the Los Alamos e-Print archive, and NCSTRL (Networked Computer Science Technical Reference Library), an online Computing Research Repository (CoRR) is being established. The Repository has been integrated into the collection of over 20,000 computer science research reports and other material available through NCSTRL (http://www.ncstrl.org) and will be linked with the ACM Digital Library. Most importantly, the Repository will be available to all members of the community at no charge. We encourage you to start using the Repository right away. For more details, see http://xxx.lanl.gov/archive/cs/intro.html. That site provides information on how to submit documents, browse, search, and subscribe to get notification of new articles of interest. Please spread the word among your colleagues and students. CoRR will only gain in value as more researchers use it. See http://www.acm.org/repository for a more detailed description of CoRR. ========================================================================== 6. New books ========================================================================== 6.1. METAMATHEMATICS OF FUZZY LOGIC Petr Hajek Trends in Logic vol. 4 Kluwer 1998, 308 pp. http:/www.wkap.nl/book.htm/0-7923-5238-6 Hardbound, USD 130.00 This book presents a systematic treatment of deductive aspects and structures of fuzzy logic understood as many valued logic {\em sui generis}. Some important systems of real-valued propositional and predicate calculus are defined and investigated. The aim is to show that fuzzy logic as a logic of imprecise (vague) propositions does have well developed formal foundations and that most things usually named ``fuzzy inference" can be naturally understood as logical deduction. There are two main groups of intended readers. First, logicians: they can see that fuzzy logic is indeed a branch of logic and may find several very interesting open problems. Second, equally important, researchers involved in fuzzy logic applications and soft computing. As a matter of fact, most of these are not professional logicians so that it can easily happen that an application, clever and succcessful as it may be, is presented in a way which is logically not entirely correct or may appear simple-minded. (Standard presentations of the logical aspects of fuzzy controllers are the most typical example.) This fact would not be very important if only the {\em bon ton} of logicians were harmed; but it is the opinion of the author (who is a mathematical logician) that a better understanding of the strictly logical basis of fuzzy logic (in the usual broad sense) is very useful for fuzzy logic appliers since if they know better what they are doing, they may hope to do it better. Still more than that: a better mutual understanding between (classical) logicians and researchers in fuzzy logic promises to lead to deeper cooperation and new results. Contents: Ch. 1 Preliminaries Ch. 2 Many-valued propositional calculi Ch. 3 Lukasiewicz propositional logic Ch. 4 Product logic, Godel logic Ch. 5 Many-valued predicate logic Ch. 6 Complexity and undecidability Ch. 7 On approximate inference Ch. 8 Generalized quantifiers and modalities Ch. 9 Miscellanea Ch. 10 Historical remarks. 6.2. "Applications of Uncertainty Formalisms" edited by Anthony Hunter and Simon Parsons has been published by Springer. The ISBN is 3-540-65312-0 and the volume number is 1455. \documentstyle{llncs} \frenchspacing \setlength{\parskip}{1em} \setlength{\parindent}{0em} \begin{document} \begin{center} {\Large {\bf Table of Contents}} \end{center} {\bf Section A: Introduction} Chapter 1: Introduction to uncertainty formalisms\\ {\it A. Hunter and S. Parsons} \dotfill 1 Chapter 2: A review of uncertainty handling formalisms\\ {\it S. Parsons and A. Hunter} \dotfill 8 {\bf Section B: Application case studies} Chapter 3: Using uncertainty management techniques in medical therapy \linebreak planning: A decision-theoretic approach\\ {\it P. Magni, R. Bellazzi and F. Locatelli} \dotfill 38 Chapter 4: An ordinal approach to the processing of fuzzy queries with flexible quantifiers\\ {\it P. Bosc, L. Li\'{e}tard and H. Prade} \dotfill 58 Chapter 5: Using uncertainty techniques in radio communication systems\\ {\it K. van Dam} \dotfill 76 Chapter 6: Handling imperfect knowledge handling in {\it Milord II} for the \linebreak identification of marine sponges \\ {\it M. Domingo, L. Godo and C. Sierra} \dotfill 88 Chapter 7: Qualitative risk assessment fulfils a need\\ {\it P. Krause, J. Fox, P. Judson and M. Patel} \dotfill 138 Chapter 8: Information retrieval and Dempster-Shafer's theory of evidence\\ {\it M. Lalmas} \dotfill 157 Chapter 9: Uncertainty measures associated with fuzzy rules for connection \linebreak admission control in ATM Networks\\ {\it M. F. N. Ramalho} \dotfill 177 Chapter 10: Handling uncertainty in control of autonmous robots\\ {\it A. Saffiotti} \dotfill 198 Chapter 11: Some problems in trying to implement uncertainty techniques in automated inspection\\ {\it D. Wilson, A. Greig, J. Gilby and R. Smith} \dotfill 225 Chapter 12: Correlation using uncertain and temporal information\\ {\it J. Bigham} \dotfill 242 Chapter 13: Arguing about beliefs and actions\\ {\it J. Fox and S. Parsons} \dotfill 266 Chapter 14: Analysis of multi-interpretable ecological monitoring information\\ {\it F. Brazier, J. Engelfriet and J. Treur} \dotfill 303 {\bf Section C: Technology for applications} Chapter 15: A local handling of inconsistent knowledge and default bases\\ {\it S. Benferhat and L. Garcia} \dotfill 325 Chapter 16: The XRay system: An implementation platform for local \linebreak query-answering in default logics\\ {\it P. Nicolas and T. Schaub} \dotfill 354 Chapter 17: Model-based diagnosis: A probabilistic extension\\ {\it A. Tawfik and E. Neufeld} \dotfill 379 Chapter 18: Background to and perspectives on possibilistic graphical models\\ {\it J. Gebhardt and R. Kruse} \dotfill 397 Chapter 19: How much does an agent believe: An extension of modal epistemic logic\\ {\it S. K. Das} \dotfill 415 Chapter 20: Safety logics\\ {\it J. Bell and Z. Huang} \dotfill 427 Chapter 21: Modelling uncertainty with propositional assumption-based systems\\ {\it R. Haenni} \dotfill 446 {\bf Author index} \dotfill 471 {\bf Subject index} \dotfill 473 \end{document} 6.3. "Classical and Fuzzy Concepts in Mathematical Logic and Applications" by M. Reghis and E. Roventa appeared in CRC Press ISBN:0-8493-3197- 6.4. Logical Structures for Representation of Knowledge and Uncertainty E. Hisdal , University of Oslo, Norway Keywords: Representation of Knowledge, Uncertainty, Artificial Intelligence (Studies in Fuzziness and Soft Computing. Ed.: J. Kacprzyk. Vol. 14) 1998 . XXIV, 419 pp. 81 figs. ISBN 3-7908-1056-8 Hardcover DM 168,- Available ========================================================================== 7. Table of Contents for Applied Intelligence Vol. 9, Number 2: Special Issue on Systems for Uncertain reasoning ========================================================================== . Applied Intelligence: Special Issue on Systems for Uncertain reasoning F. S. Correa da Silva 99 . Practical Handling of Exception-Tainted Rules and Independence Information in Possibilistic logic S. Benferhat, D. Dubois, H. Prade 101 . A Fuzzy Approach to Assessing Accident Databases P. W. H. Chung and M. Jefferson 129 . The Methofd of Assigning Incidences W. Liu, D. McBryan and A. Bundy 139 . Annoted Logic Applications for Imperfect Information D. V. Carbogin and F. S. Correa da Silva 163 . Computational Properties of Two exact Algorithms for Bayesian Networks N. L. Zhang 173 ========================================================================== 8. Conference announcements ========================================================================== 8.1. C A L L F O R P A P E R S ** U A I 99 ** THE FIFTEENTH ANNUAL CONFERENCE ON UNCERTAINTY IN ARTIFICIAL INTELLIGENCE July 30-August 1, 1999 Sweden Location to be announced (near the site of IJCAI-99) ================================= Please visit the UAI-99 WWW page at http://uai99.iet.com for the first time, the Conference on Uncertainty in Artificial Intelligence will be held outside the United States. The conference will occur immediately prior to IJCAI 99 and in close proximity to the IJCAI-99 conference site. * * * CALL FOR PAPERS Uncertainty management is a key enabling technology for the development of intelligent systems. Since 1985, the Conference on Uncertainty in Artificial Intelligence (UAI) has been the primary international forum for exchanging results on the use of principled uncertain-reasoning methods in intelligent systems. The conference has catalyzed advances in fundamental theory, efficient algorithms, and practical applications. Theory and technology first presented at UAI have been proven by wide application in the broad community, and by the success of the systems in which the technology has been embedded. The UAI Proceedings have become a basic reference for researches and practitioners who want to know about both theoretical advances and the latest applied developments in the field. The scope of UAI covers a broad spectrum of approaches to automated reasoning and decision making under uncertainty. Contributions to the proceedings address topics that advance theoretical principles or provide insights through empirical study of applications. Interests include quantitative and qualitative approaches, and traditional as well as non-classical paradigms of uncertain reasoning. Applications of automated uncertain reasoning span a broad spectrum of tasks and domains, including systems that make autonomous decisions and those designed to support human decision making through interactive use. We encourage submissions of papers for UAI '99 that report on advances in the core areas of representation, inference, learning, and knowledge acquisition, as well as on insights derived from building or using applications of uncertain reasoning. We encourage the submission of papers proposing new methodologies and tools for model construction, representation, learning, inference and experimental validation. Innovative ways to increase the expressive power and the applicability spectrum of existing methods are encouraged. Papers are welcome that present new applications of uncertain reasoning and stress the methodological aspects of their construction and use. Highlighting difficulties in existing procedures and pointing at the necessary advances in foundations and algorithms is considered an important role of presentations of applied research. Topics of interest include (but are not limited to): >> Foundations * Conceptual relationships among different uncertainty calculi * Higher order uncertainty and confidence in models * Representation of uncertainty and preference * Revision of belief and combination of information from multiple sources * Semantics of belief * Theoretical foundations of uncertain belief and decision * Uncertainty and models of causality >> Principles and Methods * Advances in diagnosis, troubleshooting, and test selection * Algorithms for reasoning and decision under uncertainty * Automated construction of inference and decision models * Combination of models from different sources * Computation and action under limited resources * Control of computational processes under uncertainty * Data structures for representation and inference * Decision making under uncertainty * Enhancing the human-computer interface with uncertain reasoning * Explanation of results of uncertain reasoning * Formal languages to represent uncertain information * Hybridization of methodologies and techniques * Integration of logic with uncertainty calculi * Markov decision processes * Methods based on probability, possibility and fuzzy logic, belief functions, rough sets, and other formalisms * Multiple agent reasoning * Planning under uncertainty * Qualitative methods and models * Reasoning at different levels of abstraction * Statistical methods for automated uncertain reasoning * Temporal reasoning * The representation and discovery of causal relationships * Time-critical decisions * Time-dependent utility * Uncertain reasoning and information retrieval * Uncertainty and methods for learning and data mining >> Empirical Study and Applications * Comparison of representation and inferential adequacy of different calculi * Empirical validation of methods for planning, learning, and diagnosis * Experience with knowledge-acquisition methods * Experimental studies of inference strategies * Methodologies for problem modeling * Nature and performance of architectures for real-time reasoning * Uncertain reasoning in embedded, situated systems (e.g., softbots) or papers focused on applications in specific domains, we suggest that the following issues be addressed in the submission: - - Why was it necessary to represent uncertainty in your domain? - - What are the distinguishing properties of the domain and problem? - - What kind of uncertainties does your application address? - - Why did you decide to use your particular uncertainty formalism? - - Which practical procedure did you follow to build the application? - - What theoretical problems, if any, did you encounter? - - What practical problems did you encounter? - - Did users/clients of your system find the results useful? - - Did your system lead to improvements in decision making? - - What approaches were effective (ineffective) in your domain? - - What methods were used to validate the effectiveness of the system? ================================== SUBMISSION AND REVIEW OF PAPERS ================================== Papers submitted for review should represent original, previously unpublished work. Papers should not be under review for presentation in any other conference. However, an extended version of the paper may be under review for publication in a scientific journal. Submitted papers will be carefully evaluated on the basis of originality, significance, technical soundness, and clarity of exposition. Papers may be accepted for presentation in plenary or poster sessions. All accepted papers will be included in the Proceedings of the Fifteenth Conference on Uncertainty in Artificial Intelligence, published by Morgan Kaufmann Publishers. An outstanding student paper will be selected for special distinction. To be considered for the outstanding student paper award, the student must be first author, and the student's advisor must certify enrollment in an academic degree program during the 1998/99 academic year. Instructions for certifying student status are available at http://uai99.iet.com/instructions.html. Authors are strongly encouraged to submit papers in the proceedings format. Format information is available at http://uai99.iet.com/instructions.html. Submitted papers must be no more than ten pages in proceedings format, (about 7000 words). Accepted papers will be limited to 8 pages with 2 additional pages for a fee. We strongly encourage the electronic submission of papers. To submit a paper electronically, send an electronic version of the paper (Postscript format) to the following address: uai99@iet.com The subject line of this message should be: $.ps, where $ is an identifier created from the last name of the first author, followed by the first initial of the author's first name. Multiple submissions by the same first author should be indicated by adding a number (e.g., pearlj2.ps) to the end of the identifier. Additionally, the paper abstract and data should be sent by using the electronic form at the following address: http://uai99.iet.com/data.html Authors unable to submit papers electronically should send 5 copies of the complete paper to one of the Program Chairs at the addresses listed below. Authors unable to use the electronic form to submit the abstract should provide the following information (by sending a message to the e-mail address above): * Paper title (plain text) * Author names, including student status (plain text) * Surface mail address, e-mail address, and voice phone number for a contact author (plain text) * A short abstract including keywords (plain text) * Primary and secondary classification indices selected from conference topics listed above. * Indicate the preferred type of presentation: poster or plenary ++++++++++++++++++++++++++++++ Important Dates ++++++++++++++++++++++++++++++ ** All submissions must be received by: Sunday, February 21, 1999 ** Notification of acceptance on or before: Monday, April 12, 1999 ** Camera-ready copy due: Friday, May 7, 1999 ** Conference dates: Friday-Sunday, July 30 - August 1, 1999 ** Full day course on Uncertain Reasoning: Thursday, July 29, 1999 Conference E-mail Address ========================= Please send all inquiries (submissions and conference organization) to the following e-mail address: uai99@iet.com Program Co-chairs: Kathryn Blackmond Laskey Department of Systems Engineering and Operations Research George Mason University airfax, VA 22030-4444 USA Phone: (703) 993-1644 fax: (703) 993-1706 E-mail: klaskey@gmu.edu Henri Prade I.R.I.T. Universit=E9 Paul Sabatier 118 route de Narbonne 31062 Toulouse Cedex 4 FRANCE Phone: (33) 561 55 6579 fax : (33) 561 55 6239 E-mail: uai99@irit.fr General Conference Chair: Gregory F. Cooper Center for Biomedical Informatics University of Pittsburgh Suite 8084 Forbes Tower 200 Lothrop Street Pittsburgh, PA 15213-2582 USA Phone: (412) 647-7113 fax: (412) 647-7190 E-mail: gfc@cbmi.upmc.edu Refer to the UAI-98 WWW home page for late-breaking information: http://uai99.iet.com/ ------------------------------------------------------------------ 8.2. SECOND CALL FOR PAPERS FOR ECSQARU'99 ........................................................... European Conference on Symbolic and Quantitative Approaches to Reasoning with Uncertainty ........................................................... 5-9 July 1999 at UCL, London, UK Webpage at www.cs.ucl.ac.uk/staff/a.hunter/ecsqaru AIMS AND SCOPE: Uncertainty is in an increasingly important research topic in many areas of computer science. The main European forum for the subject is the European Conference on Symbolic and Quantitative Approaches to Reasoning and Uncertainty (ECSQARU). These have been held in Marsellies (1991), Granada (1993), Fribourg (1995), and Bonn (1997). The next in the series is ECSQARU'99 in London in July 1999. AREAS FOR CONTRIBUTION (not exclusive): Default reasoning; Belief revision; Logics for reasoning with uncertainty; Paraconsistent logics; Belief functions; Bayesian networks; Probabilistic reasoning; Fuzzy systems; Aggregation of arguments; Inconsistency handling; Decision systems; Fusion systems; Argumentation systems; Applications of uncertainty formalisms; Automated reasoning systems for uncertainty formalisms; Machine learning for uncertainty formalisms. PROGRAM COMMITTEE: Tony Hunter (London) - Program chair Henri Prade (Toulouse) - Data fusion Finn Jensen (Aalborg) - Bayesian networks Torsten Schaub (Potsdam) - Default systems Philippe Smets (Bruxelles) - Belief functions Dov Gabbay (London) - Logics Rudolf Kruse (Magdeburg) - Fuzzy methods SUBMISSION OF PAPERS: Please limit submissions to a maximum of 10 pages, preferrably in LNCS format. Details on the LNCS format, including a Latex .sty file, can be obtained via the conference webpage. To submit a paper, please send it as a postscript file by email to Tony Hunter at a.hunter@cs.ucl.ac.uk, or post four copies of it, to Tony Hunter at the Department of Computer Science, University College London, Gower Street, London WC1E 6BT, UK. The proceedings will be published in the LNCS Series IMPORTANT DATES: Submission deadline 31 January 1999; Notification of acceptance 12 March 1999; CRC for accepted papers 16 April 1999; Workshops and tutorials 5-6 July 1999; Main conference 7-9 July 1999 8.3. ISIPTA '99 THE FIRST INTERNATIONAL SYMPOSIUM ON IMPRECISE PROBABILITIES AND THEIR APPLICATIONS Ghent, Belgium 30 June - 2 July 1999 CALL FOR PAPERS ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ Imprecise probability is a generic term for the many mathematical models which measure chance or uncertainty without sharp numerical probabilities. These include models such as belief functions, Choquet capacities, comparative probability orderings, convex sets of probability measures, fuzzy measures, interval-valued probabilities, possibility measures, and upper and lower expectations or previsions. Such models are needed in inference problems where the relevant information is scarce, vague or conflicting, and in decision problems where preferences are incomplete. Imprecise probability models are now being studied and applied by a large number of researchers, working in a great variety of fields: statistics, artificial intelligence, economics, psychology, engineering, medicine, management science, etc. There is too little communication between people working in different, but related, fields or in different countries, and this sometimes results in duplication of research or in inconsistent terminology. We believe, therefore, that there is a need for a regular symposium on imprecise probabilities. Symposium location and dates ---------------------------- The first International Symposium on Imprecise Probabilities and Their Applications (ISIPTA ‘99) will be held at the Universiteit Gent, in Ghent, Belgium, 30 June - 2 July 1999. Ghent is one of the oldest cities in Flanders, the Dutch-speaking northern part of Belgium. It is a place rich in history, architecture, culture and gastronomy. It is also very close to other interesting historical Flemish cities (Bruges, Antwerp). Links to more tourist information about Ghent, Flanders and Belgium can be found at the symposium web site: http://ensmain.rug.ac.be/~isipta99. Organising Committee -------------------- Chairman: Gert de Cooman Members: Dirk Aeyels, Fabio G. Cozman, Heidi Jansegers, Etienne E. Kerre, Serafín Moral, Da Ruan, Bartel Van de Walle, Peter Walley. Program Committee ----------------- Gert de Cooman (Belgium) Luis R. Pericchi (Venezuela) Fabio G. Cozman (Brazil) Henri Prade (France) Didier Dubois (France) David Schmeidler (Israel, USA) Angelo Gilio (Italy) Teddy Seidenfeld (USA) Michel Grabisch (France) Philippe Smets (Belgium) Jean-Yves Jaffray (France) Michael Smithson (Australia) Brigitte Jaumard (Canada) Wynn Stirling (USA) George J. Klir (USA) Lev V. Utkin (Russia) Serafín Moral (Spain) Peter Walley (Chairman, Australia) Robert Nau (USA) Nic Wilson (UK) Themes of the symposium ----------------------- The first symposium will emphasise two main themes: (i) unifying concepts and relationships between the theories: This concerns the connections between the various mathematical theories and the possibility of achieving some kind of unification or synthesis. (ii) applications to other fields: Important work has already been done in statistics, economics, finance, experimental psychology, artificial intelligence and expert systems, engineering, systems theory, reliability, robotics and other fields, and it is hoped that some of this work will be surveyed at the symposium. Symposium language ------------------ The working language of the symposium will be English. No simultaneous translation in other languages will be available. Contributions ------------- Those wishing to present a paper at the symposium should submit a short paper of 4 to 10 pages by 31 January 1999. We expect electronic submissions, in Postscript or PDF format. Papers should be sent to the symposium e-mail address: isipta99@ensmain.rug.ac.be. Please see the symposium web site http://ensmain.rug.ac.be/~isipta99 for instructions about how to submit papers. We encourage papers on the two main themes of the symposium: unifying concepts and the possibility of a unified theory; and applications. In particular, we encourage but will not limit the contributions to: (i) papers that are of widespread and general interest to people working with imprecise probabilities (ii) surveys of specific fields of application or potential application of imprecise probabilities (iii) surveys of work on specific topics such as conditioning, independence and coherence (iv) surveys of recent developments in the theory of imprecise probabilities. The Program Committee will decide which papers are accepted. The successful authors will be invited to submit a final, possibly extended, version of their paper after the symposium, for publication in a volume of symposium proceedings. All the papers that are accepted for the symposium will be made available on the symposium web site http://ensmain.rug.ac.be/~isipta99 well before the symposium. Important dates --------------- Submission deadline: 31 January 1999 Notification of acceptance: 31 March 1999 Deadline for revised papers: 30 April 1999 Deadline for early registration: 30 April 1999 Symposium: 30 June - 2 July 1999 Questions --------- If you have any questions about the symposium, please contact the ISIPTA ’99 Secretariat at the address given below. Electronic preregistration -------------------------- If you want to be kept informed about the current symposium or later symposia, you can fill out the electronic preregistration form at the symposium web site. The Imprecise Probabilities Project (IPP) ----------------------------------------- The web site http://ensmain.rug.ac.be/~ipp contains some introductory articles about imprecise probabilities, plus an extensive (and still growing) bibliography, and it will soon contain a collection of survey articles on special types of imprecise probability models. The Imprecise Probabilities Project also maintains a repository of abstracts of papers on imprecise probabilities, and an electronic mailing list. More information about these services can be found on the IPP web site. ISIPTA '99 Secretariat ---------------------- Address: ISIPTA ’99 Secretariat p/a Gert de Cooman Universiteit Gent Onderzoeksgroep SYSTeMS Technologiepark 9 9052 Zwijnaarde Belgium Telephone: +32-(0)9-264.56.53 Fax: +32-(0)9-264.58.40 E-mail: isipta99@ensmain.rug.ac.be Symposium Web Site ------------------ http://ensmain.rug.ac.be/~isipta99 IPP Web Site ------------ http://ensmain.rug.ac.be/~ipp