http://repub.eur.nl/res/aut/40/ List of Publications en http://repub.eur.nl/static-eur/img/logo.png http://repub.eur.nl http://repub.eur.nl/res/pub/1929/ 2005-04-03T00:00:00Z Artificial stock markets are built with diffuse priors in mind regarding trading strategies and price formation mechanisms. Diffuse priors are a natural consequence of the unknown relation between the various elements that drive market dynamics and the large variety of market organizations, findings, however, might hold only within the specific market settings. In this paper we propose a framework for building agent-based artificial stock markets. We present the mechanism of the framework based on a previously identified list of organizational and behavioural aspects. Within the framework experiments with arbitrary many trading strategies, acting in various market organizations can be conducted in a flexible way, without changing its architecture. In this way experiments of other artificial stock markets, as well as theoretical models can be replicated and their findings compared. Comparisons of the different experimental results might indicate whether findings are due to traders’ behaviour or to the chosen market structure and could suggest how to improve market quality. http://repub.eur.nl/res/pub/1900/ 2005-02-18T00:00:00Z Artificial stock markets are designed with the aim to study and understand market dynamics by representing (part of) real stock markets. Since there is a large variety of real stock markets with several partially observable elements and hidden processes, artificial markets differ regarding their structure and implementation. In this paper we analyze to what degree current artificial stock markets reflect the workings of real stock markets. In order to conduct this analysis we set up a list of factors which influence market dynamics and are as a consequence important to consider for designing market models. We differentiate two categories of factors: general, well-defined aspects that characterize the organization of a market and hidden aspects that characterize the functioning of the markets and the behaviour of the traders. http://repub.eur.nl/res/pub/459/ 2003-06-26T00:00:00Z This paper deals with algorithms searching trees generated by two-person, zero-sum games with perfect information. The standard algorithm in this field is alpha-beta. We will discuss this algorithm as well as extensions, like transposition tables, iterative deepening and NegaScout. Special attention is devoted to domain knowledge pertaining to game trees, more specifically to solution trees. The above mentioned algorithms implement depth first search. The alternative is best first search. The best known algorithm in this area is Stockman's SSS*. We treat a variant equivalent to SSS* called SSS-2. These algorithms are provably better than alpha-beta, but it needs a lot of tweaking to show this in practice. A variant of SSS-2, cast in alpha-beta terms, will be discussed which does realize this potential. This algorithm is however still worse than NegaScout. On the other hand, applying a similar idea as the one behind NegaScout to this last SSS version yields the best (sequential) game tree searcher known up till now: MTD(f). http://repub.eur.nl/res/pub/464/ 2003-06-26T00:00:00Z This report contains a non-formal motivation and description of ADL-d, a graphical design technique for parallel and distributed software. ADL-d allows a developer to construct an application in terms of communicating processes. The technique distinguishes itself from others by its use of highly orthogonal concepts, and support for automated code generation. Without being committed to one particular design method, ADL-d as a technique can be used from the early phases of application design through phases that concentrate on algorithmic design, and final implementation on some target platform. In this report, we discuss and motivate all ADL-d components, including recently incorporated features such as support for connection-oriented communication, support for modeling dynamically changing communication structures, and a formal semantical basis for each ADL-d component. Also, we discuss our ADL-d implementation, and place ADL-d in context by discussing some related work. http://repub.eur.nl/res/pub/14408/ 2003-04-01T00:00:00Z In this paper, we describe a Multi-Agent System which is capable of finding a feasible solution of a class of distributed problems, in which the subproblems share a single a sum constraint. Emphasis is given to correctness issues and termination detection. http://repub.eur.nl/res/pub/161/ 2002-01-31T00:00:00Z By using Paradigm, it is possible to model cooperating processes and to make the communication between these processes very clear. This report gives a formal description of this modeling method using state-transition diagrams in order to model processes and homomorphismes and interleavings in order to model the cooperation and synchronization of the processes involved. Paradigm has been used succesfully as the modeling language of Socca, a software proces modeling method. http://repub.eur.nl/res/pub/7722/ 1999-05-26T00:00:00Z In this paper we will describe a Multi-Agent System which is capable of finding a feasible solution of a specially structured linear programming problem. Emphasis is given to correctness issues and termination detection. http://repub.eur.nl/res/pub/763/ 1998-01-01T00:00:00Z In this paper, a theory of game tree algorithms is presented, entirely based upon the concept of solution tree. Two types of solution trees are distinguished: max and min trees. Every game tree algorithm tries to prune nodes as many as possible from the game tree. A cut-off criterion in terms of solution trees will be formulated, which can be used to eliminate nodes from the search without affecting the result. Further, we show that any algorithm actually constructs a superposition of a max and a min solution tree. Finally, we will see, how solution trees and the related cutoff criterion are applied in major game tree algorithms, like alpha-beta and MTD. http://repub.eur.nl/res/pub/520/ 1997-01-01T00:00:00Z We consider dynamic Kahn-like dataflow networks defined by a simple language L containing the fork-statement. The first part of the Kahn principle states that such networks are deterministic on the I/O level: for each network, different executions provided with the same input deliver the same output. The second part of the principle states that the function from input streams to output streams (which is now defined because of the first part) can be obtained as a fixed point of a suitable operator derived from the network specification. The first part has been proven by us in an earlier publication. To prove the second part, we will use the metric framework. We introduce a nondeterministic transition system NT from which we derive an operational semantics On. We also define a deterministic transition system DT and prove that the operational semantics Od derived from DT is the same as On. Finally, we define a denotational semantics D and prove D = Od. This implies On = http://repub.eur.nl/res/pub/468/ 1996-01-01T00:00:00Z In this paper a complete theory of game tree algorithms is presented, entirely based upon the notion of a solution tree. Two types of solution trees are distinguished: max and min solution trees respectively. We show that most game tree algorithms construct a superposition of a max and a min solution tree. Moreover, we formulate a general cut-off criterion in terms of solution trees. In the second half of this paper four well known algorithms, viz., alphabeta, SSS*, MTD and Scout are studied extensively. We show how solution trees feature in these algorithms and how the cut-off criterion is applied. http://repub.eur.nl/res/pub/511/ 1996-01-01T00:00:00Z Branch and bound algorithms can be characterized by a small set of basic rules that are applied in a divide-and-conquer-like framework. The framework is about the same in all applications, whereas the specification of the rules is problem dependent. Building a framework is a rather simple task in sequential implementations, but must not be underestimated in the parallel case, especially if an efficient branch and bound algorithm is required. In generic branch and bound models, the basic rules can be clearly identified within the framework, and, hence, it can be developed independently from the application. Furthermore, it gives the user the opportunity to concentrate on the actual problem to be solved, without being distracted by user-irrelevant issues like the properties of the underlying architecture. In this paper, we will discuss an object oriented approach to generic branch and bound. We will show how object orientation can help us to build a flexible branch and bound framework, that is able to perform like any branch and bound algorithm that fits into some powerful taxonomies known from the literature. We will define an interface for the specification of the problem dependent parts, and we will give a first indication of how the user can tune the framework if a non-default behavior is desired. http://repub.eur.nl/res/pub/1438/ 1995-01-01T00:00:00Z The parallel execution of branch and bound algorithms can result in seemingly unreasonable speedups or slowdowns. Almost never the speedup is equal to the increase in computing power. For synchronous parallel branch and bound, these effects have been studiedd extensively. For asynchronous parallelizations, only little is known. In this paper, we derive sufficient conditions to guarantee that an asynchronous parallel branch and bound algorithm (with elimination by lower bound tests and dominance) will be at least as fast as its sequential counterpart. The technique used for obtaining the results seems to be more generally applicable. The essential observations are that, under certain conditions, the parallel algorithm will always work on at least one node, that is branched from by the sequential algorithm, and that the parallel algorithm, after elimination of all such nodes, is able to conclude that the optimal solution has been found. Finally, some of the theoretical results are brought into connection with a few practical experiments. http://repub.eur.nl/res/pub/1439/ 1995-01-01T00:00:00Z Many (parallel) branch and bound algorithms look very different from each other at first glance. They exploit, however, the same underlying computational model. This phenomenon can be used to define branch and bound algorithms in terms of a set of basic rules that are applied in a specific (predefined) order. In the sequential case, the specification of Mitten's rules turns out to be sufficient for the development of branch and bound algorithms. In the parallel case, the situation is a bit more complicated. We have to consider extra parameters such as work distribution and knowledge sharing. Here, the implementation of parallel branch and bound algorithms can be seen as a tuning of the parameters combined with the specification of Mitten's rules. These observations lead to generic systems, where the user provides the specifications of the problem to be solved, and the system generates a branch and bound algorithm running on a specific architecture. We will discuss some proposals that appeared in the literature. Next, we raise the question whether the proposed models are flexible enough. We analyze the design decisions to be taken when implementing a parallel branch and bound algorithm. It results in a classification model, which is validated by checking whether it captures existing branch and bound implementations. Finally, we return to the issue of flexibility of existing systems, and propose to add an abstract machine model to the generic framework. The model defines a virtual parallel branch and bound machine, within which the design decisions can be expressed in terms of the abstract machine. We will outline some ideas on which the machine may be based, and present directions of future work. http://repub.eur.nl/res/pub/1440/ 1995-01-01T00:00:00Z This paper introduces a new paradigm for minimax game-tree search algorithms. MT is a memory-enhanced version of Pearl's Test procedure. By changing the way MT is called, a number of best-first game-tree search algorithms can be simply and elegantly constructed (including SSS*). Most of the assessments of minimax search algorithms have been based on simulations. However, these simulations generally do not address two of the key ingredients of high performance game-playing programs: iterative deepening and memory usage. This paper presents experimental data from three game-playing programs (checkers, Othello and chess), covering the range from low to high branching factor. The improved move ordering due to iterative deepening and memory usage results in significantly different results from those portrayed in the literature. Whereas some simulations show alpha-beta expanding almost 100% more leaf nodes than other algorithms [Marsland, Reinefeld & Schaeffer, 1987], our results showed variations of less than 20%. One new instance of our framework MTD(f) out-performs our best alpha-beta searcher (aspiration NegaScout) on leaf nodes, total nodes and execution time. To our knowledge, these are the first reported results that compare both depth-first and best-first algorithms given the same amount of memory. http://repub.eur.nl/res/pub/1441/ 1995-01-01T00:00:00Z In 1979 Stockman introduced the SSS* minimax search algorithm that dominates alpha-beta in the number of leaf nodes expanded. Further investigation of the algorithm showed that it had three serious drawbacks, which prevented its use by practitioners: it is difficult to understand, it has large memory requirements, and it is slow. This paper presents an alternate formulation of SSS*, in which it is implemented as a series of alpha-beta calls that use a transposition table (ABSSS). The reformulation solves all three perceived drawbacks of SSS*, making it a practical algorithm. Further, because the search is now based on alpha-beta, the extensive research on minimax search enhancements can be easily integrated into ABSSS. To test ABSSS in practise, it has been implemented in three state-of-the-art programs: for checkers, Othello and chess. ABSSS is comparable in performance to alpha-beta on leaf node count in all three games, making it a viable alternative to alpha-beta in practise. Whereas SSS* has usually been regarded as being entirely different from alpha-beta, it turns out to be just an alpha-beta enhancement, like null-window searching. This runs counter to published simulation results. Our research leads to the surprising result that iterative deepening versions of alpha-beta can expand fewer leaf nodes than iterative deepening versions of SSS* due to dynamic move re-ordering. http://repub.eur.nl/res/pub/1455/ 1994-01-01T00:00:00Z We consider dynamic Kahn-like data flow networks, i.e. networks consisting of deterministic processes each of which is able to expand into a subnetwork. The Kahn principle states that such networks are deterministic, i.e. that for each network we have that each execution provided with the same input delivers the same output. Moreover, the principle states that the output streams of such networks can be obtained as the smallest fixed point of a suitable operator derived from the network specification. This paper is meant as a first step towards a proof of this principle. For a specific subclass of dynamic networks, linear arrays of processes, we define a transition system yielding an operational semantics which defines the meaning of a net as the set of all possible interleaved executions. We then prove that, although on the execution level there is much nondeterminism, this nondeterminism disappears when viewing the system as a transformation from an input stream to an output stream. This result is obtained from the graph of all computations. For any configuration such a graph can be constructed. All computation sequences that start from this configuration and that are generated by the operational semantics are embedded in it. http://repub.eur.nl/res/pub/1456/ 1994-01-01T00:00:00Z A game tree algorithm is an algorithm computing the minimax value of the root of a game tree. Many algorithms use the notion of establishing proofs that this value lies above or below some boundary value. We show that this amounts to the construction of a solution tree. We discuss the role of solution trees and critical trees in the following algorithms: Principal Variation Search, alpha-beta, and SSS-2. A general procedure for the construction of a solution tree, based on alpha-beta and Null-Window-Search, is given. Furthermore two new examples of solution tree-based algorithms are presented, that surpass alpha-beta, i.e., never visit more nodes than alpha-beta, and often less. http://repub.eur.nl/res/pub/1466/ 1993-01-01T00:00:00Z A unifying framework for game tree algorithms is GSEARCH, designed by Ibaraki. In general, a relatively great deal of memory is necessary for instances of this framework. Another framework from Ibaraki is RSEARCH, in which the use of memory can be controlled. In this paper variants of above frameworks are introduced, to be called Gsearch and Rsearch respectively. It is shown that, in these frameworks, the classical alpha-beta algorithm is the depth-first search instance and H* is a best first search instance. Furthermore two new algorithms, Maxsearch and Minsearch, are presented, both as best-first search instances. Maxsearch is close to SSS* and SSS-2, whereas Minsearch is close to dual SSS*. http://repub.eur.nl/res/pub/1490/ 1992-01-01T00:00:00Z Well-known algorithms for the evaluation of the minimax function in game trees are alpha-beta and SSS*. An improved version of SSS* is SSS-2. All these algorithms don't use any heuristic information on the game tree. In this paper the use of heuristic information is introduced into the alpha-beta and the SSS-2 algorithm. Extended versions of these algorithms are presented. The subset of nodes which is visited during execution of each algorithm is characterised completely. http://repub.eur.nl/res/pub/1491/ 1992-01-01T00:00:00Z In this paper we present a classification of parallel branch and bound algorithms, and elaborate on the consequences of particular parameter settings. The taxonomy is based upon how the algorithms handle the knowledge about the problem instance to be solved, generated during execution. The starting point of the taxonomy is the generally accepted description of the sequential branch and bound algorithm, as presented in, for example, [Mitten 1970] and [Ibaraki 1976a, 1976b, 1977a, 1977b]. http://repub.eur.nl/res/pub/1502/ 1990-01-01T00:00:00Z We show that the technique to prove equivalence of operational and denotational cpo based semantics using retractions, as introduced in de Bruin & Vink [1989] for a sequential backtracking language, can be applied to parallel languages as well. We prove equivalence for a uniform language in which procedure calls need not be guarded. The unguardedness is taken care of by giving a semantics in which the nondeterminism is demonic. http://repub.eur.nl/res/pub/1503/ 1990-01-01T00:00:00Z A new version of the SSS* algorithm for searching game trees is presented. This algorithm is built around two recursive procedures. It finds the minimax value of a game tree by first establishing an upper bound to this value and then successively trying in a top down fashion to tighten this bound until the minimax value has been obtained. This approach has several advantages, most notably that the algorithm is more perspicuous. Correctness and several other properties of SSS* can now more easily be proven. As an example we prove Pearl's characterization of the nodes visited by SSS*. Finally the new algorithm is transformed into a practical version, which allows an efficient use of memory. http://repub.eur.nl/res/pub/1505/ 1989-01-01T00:00:00Z We present an operational model O and a continuation based denotational model D for a uniform variant of PROLOG, including the cut operator. The two semantical definitions make use of higher order transformations Phi and Psi, respectively. We prove O and D equivalent in a novel way by comparing yet another pair of higher order transformations Phi~ and Psi~, that yield Phi and Psi, respectively, by application of a suitable abstraction operator. http://repub.eur.nl/res/pub/1507/ 1989-01-01T00:00:00Z We present a denotational continuation semantics for PROLOG with cut. First a uniform language B is studied, which captures the control flow aspects of PROLOG. The denotational semantics for B is proven equivalent to a transition system based operational semantics. The congruence proof relies on the representation of the operational semantics as a chain of approximations and on a convenient induction principle. Finally, we interpret the abstract language B such that we obtain equivalent denotational and operational models for PROLOG itself. http://repub.eur.nl/res/pub/1510/ 1987-01-01T00:00:00Z Parallel computation offers a challenging opportunity to speed up the time consuming enumerative procedures that are necessary to solve hard combinatorial problems. Theoretical analysis of such a parallel branch and bound algorithm is very hard and empirical analysis is not straightforward because the performance of a parallel algorithm cannot be evaluated simply by executing the algorithm on a few parallel systems. Among the difficulties encountered are the noise produced by other users on the system, the limited variation in parallelism (the number of processors in the system is strictly bounded) and the waste of resources involved: most of the time, the outcomes of all computations are already known and the only issue of interest is when these outcomes are produced. We will describe a way to simulate the execution of parallel branch and bound algorithms on arbitrary parallel systems in such a way that the memory and cpu requirements are very reasonable. The use of simulation has only minor consequences for the formulation of the algorithm.