This text analyses the papers accepted for the workshop "Reuse of designs: an interdisciplinary cognitive approach". Several dimensions and questions considered as important (by the authors and/or by us) are addressed: What about the "interdisciplinary cognitive" character of the approaches adopted by the authors? Is design indeed a domain where the use of CBR is particularly suitable? Are there important distinctions between CBR and other approaches? Which types of knowledge -other than cases- is being, or might be, used in CBR systems? With respect to cases: are there different "types" of case and different types of case use? which formats are adopted for their representation? do cases have "components"? how are cases organised in the case memory? Concerning their retrieval: which types of index are used? on which types of relation is retrieval based? how does one retrieve only a selected number of cases, i.e., how does one retrieve only the "best" cases? which processes and strategies are used, by the system and by its user? Finally, some important aspects of CBR system development are shortly discussed: should CBR systems be assistance or autonomous systems? how can case knowledge be "acquired"? what about the empirical evaluation of CBR systems? The conclusion points out some lacking points: not much attention is paid to the user, and few papers have indeed adopted an interdisciplinary cognitive approach.

In this paper we experiment with a 2-player strategy board game where playing models are evolved using reinforcement learning and neural networks. The models are evolved to speed up automatic game development based on human involvement at varying levels of sophistication and density when compared to fully autonomous playing. The experimental results suggest a clear and measurable association between the ability to win games and the ability to do that fast, while at the same time demonstrating that there is a minimum level of human involvement beyond which no learning really occurs.

We investigate systematically the impact of human intervention in the training of computer players in a strategy board game. In that game, computer players utilise reinforcement learning with neural networks for evolving th eir playing strategies and demonstrate a slow learning speed. Human intervention can significan tly enhance learning performance, but carrying it out systematically seems to be more of a problem of an integrated game development environment as opposed to automatic evolutionary learning.

Pattern classification is an important task in several image processing, statistical learning, and data mining applications. The most popular pattern classifiers are Bayesian classifiers. There are many well known methods for represent ing Bayesian classifiers, but one of the most useful method is by discriminant functions . These functions provide inter-class decision surfaces for Bayesian classifier s. Discriminant functions assume several forms depending on the probability density of the feature space. But most attention has been received by discriminant functions that assume multivariate Gaussian distribution [1].

This paper addresses the issue of policy evaluation in Markov Decision Processes, using linear function approximation. It provides a unified view of algorithms such as TD(lambda), LSTD(lambda), iLSTD, residual-gradient TD. It is asserted that they all consist in minimizing a gradient function and differ by the form of this function and their means of minimizing it. Two new schemes are introduced in that framework: Full-gradient TD which uses a generalization of the principle introduced in iLSTD, and EGD TD, which reduces the gradient by successive equi-gradient descents. These three algorithms form a new intermediate family with the interesting property of making much better use of the samples than TD while keeping a gradient descent scheme, which is useful for complexity issues and optimistic policy iteration.

Constraint Programming (CP)[21] is a powerful technique for spec ifying Constraint Satisfaction Problems (CSPs) based on allowing a constraintprogrammer to model problems in terms of high-level constraints. Using such global constraints allows easier specification of problems but also allows for faster solve rs that take advantage of the structure in the problem. The classica l approach to CSP solving is to explore the search tree of all possible assignment s to the variables in a depth-first search backtracking manner, guided by v arious heuristics, until a solution is found or proven not to exist. One of the most basic techniques for reducing the number of search tree nodes explore d is to perform domain propagation at each node. In order to get as much domain propagation as possible we wish for each constraint to remove from the variable d omains all values that cannot participate in a solution to that constraint.

The local reconstruction of a railway schedule following a small perturbation of the traffic, seeking minimization of th e total accumulated delay, is a very difficult and tightly constrained combinatorial problem. Notoriously enough, the railway company's public image degrades proportionally to the amount of daily delays, and the same goes for its profit! This paper describes an inoculation procedure which greatly enhances an evolutionary algorithm for train re-schedulin g. The procedure consists in building the initial population around a pre-computed solution based on problem-related information available beforehand. The optimization is performed by adapting times of departure and arrival, as well as allocation of tracks, for eac h train at each station. This is achieved by a permutation-based evolutionary algorithm that relies on a semi-greedy heuristic scheduler to gradually reconstruct the schedule by inserting trains one after another. Experimental results are presented on various instances of a large real-world case involving around 500 trains and more than 1 million constraints. In terms of competition with commercial mathematical programming tool ILOG CPLEX, it appears that within a large class of instances, excluding trivial instances as well as too difficult ones, and with very few exceptions, a clever initialization turns an encouragi ng failure into a clear-cut success auguring of substantial fin an-cial savings.

This approach will enab le robots to have complex responses to unfamiliar situations without the need for e ither a computationally intensive central processor or preprogrammed prior antic ipation of all possible situations. Conventional robots achieve adaptive behavi or by either digital programmed world-models (Bekey, 2005) or through large numbers of finite state machines programmed for small tasks - sensor input/actuator output (Arkin, 1999). The former approach requires massive amounts of up-front programming and re sults in a brittle computational system. New and/or unexpected events will result in r obot behavior not necessarily appropriate to the situation since the robot can only draw from a limited library of preprogrammed behaviors. The latter approach has the a dvantage of not requiri ng a world model but suffers from the same problem of not re sponding appropriately in many situations.

We develop a new collaborative filtering (CF) method that combines both previously known users' preferences, i.e. standard CF, as well as product/user attributes, i.e. classical function approximation, to predict a given user's interest in a particular product. Our method is a generalized low rank matrix completion problem, where we learn a function whose inputs are pairs of vectors -- the standard low rank matrix completion problem being a special case where the inputs to the function are the row and column indices of the matrix. We solve this generalized matrix completion problem using tensor product kernels for which we also formally generalize standard kernel properties. Benchmark experiments on movie ratings show the advantages of our generalized matrix completion method over the standard matrix completion one with no information about movies or people, as well as over standard multi-task or single task learning methods.

Outsourcing of complex IT infrastructure to IT service providers has increased substantially during the past years. IT service providers must be able to fulfil their service-quality commitments based upon predefined Service Level Agreements (SLAs) with the service customer. They need to manage, execute and maintain thousands of SLAs for different customers and different types of services, which needs new levels of flexibility and automation not available with the current technology. The complexity of contractual logic in SLAs requires new forms of knowledge representation to automatically draw inferences and execute contractual agreements. A logic-based approach provides several advantages including automated rule chaining allowing for compact knowledge representation as well as flexibility to adapt to rapidly changing business requirements. We suggest adequate logical formalisms for representation and enforcement of SLA rules and describe a proof-of-concept implementation. The article describes selected formalisms of the ContractLog KR and their adequacy for automated SLA management and presents results of experiments to demonstrate flexibility and scalability of the approach.