Dominique Bouthinon

Collaborative Concept Learning: Non Individualistic vs Individualistic Agents

This article addresses collaborative learning in a multi-agent system: each agent revises incrementally its beliefs B (a concept representation) to keep it consistent with the whole set of information K (the examples) that he has received from the environment or other agents. In SMILE this notion of consistency was extended to a group of agents and a unique consistent concept representation was so maintained inside the group. In the present paper, we present iSMILE in which the agents still provide examples to other agents but keep their own concept representation. We will see that iSMILE is more time consuming and loses part of its learning ability, but that when agents cooperate at classification time, the group benefits from the advantages of ensemble learning.

Concept Learning from (Very) Ambiguous Examples

We investigate here concept learning from incomplete examples, denoted here as ambiguous . We start from the learning from interpretations setting introduced by L. De Raedt and then follow the informal ideas presented by H. Hirsh to extend the Version space paradigm to incomplete data: a hypothesis has to be compatible with all pieces of information provided regarding the examples. We propose and experiment an algorithm that given a set of ambiguous examples, learn a concept as an existential monotone DNF. We show that 1) boolean concepts can be learned, even with very high incompleteness level as long as enough information is provided, and 2) monotone, non monotone DNF (i.e. including negative literals), and attribute-value hypotheses can be learned that way, using an appropriate background knowledge. We also show that a clever implementation, based on a multi-table representation is necessary to apply the method with high levels of incompleteness.

Formal Concept Analysis of Attributed Networks

We consider attribute pattern mining in an attributed graph through recent developments of Formal Concept Analysis. The core idea is to restrain the extensional space, i.e., the space of possible pattern extensions in the vertex set O, to vertex subsets satisfying some topological property. We consider two levels. At the abstract level, we reduce the extension of each pattern in such a way that the corresponding abstract extension induces a subgraph whose nodes satisfy some connectivity property. At the local level a pattern has various extensions each associated with a connected component of the abstract subgraph associated with the pattern. We obtain that way abstract closed patterns and local closed patterns, together with abstract and local implications. Furthermore, working at abstract and local levels leads to proper interestingness measures that evaluate to what extent patterns and implications are related to the topological information. Finally, we relate local concepts to network communities and show that to plainly express such a notion it may be necessary to apply our methodology to a new graph derived from the original network. We consider in particular the detection and ordering of k-communities in subgraphs of an attributed network.

Abstract and Local Rule Learning in Attributed Networks

We address the problem of finding local patterns and related local knowledge, represented as implication rules, in an attributed graph. Our approach consists in extending frequent closed pattern mining to the case in which the set of objects is the set of vertices of a graph, typically representing a social network. We recall the definition of abstract closed patterns, obtained by restricting the support set of an attribute pattern to vertices satisfying some connectivity constraint, and propose a specificity measure of abstract closed patterns together with an informativity measure of the associated abstract implication rules. We define in the same way local closed patterns, i.e. maximal attribute patterns each associated to a connected component of the subgraph induced by the support set of some pattern, and also define specificity of local closed patterns together with informativity of associated local implication rules. We also show how, by considering a derived graph, we may apply the same ideas to the discovery of local patterns and local implication rules in non disjoint parts of a subgraph as k-cliques communities.

Local Knowledge Discovery in Attributed Graphs

We address the problem of finding local patterns and related local knowledge in an attributed graph. Our approach consists in extending the methodology of frequent closed pattern mining to the case in which the set of objects, in which are to be found the patterns support sets, is the set of vertices of a graph, typically representing a social network. We propose an algorithm to enumerate triples (c,e,l) where c is a (global) closed pattern which leads in the region e of the graph to a local closed pattern l and define a basis of implication rules expressing what new attributes l\c appear when focussing in this region. We discuss how to apply this methodology to the detection of frequent k-communities.

Bi-Pattern Mining of Two Mode and Directed Networks.

In two-mode networks there are two kinds of vertices, i.e objects, each being possibly described with a proper attribute set. This means that to select a subnetwork according to vertex descriptions we have to consider a pair of vertex subsets. A common technique is to extract from a network an essential subnetwork, the core subgraph of the network. Formal Concept Analysis and closed pattern mining were previously applied to networks with the purpose of reducing extensions of patterns to be core subgraphs. To apply this methodology to two-mode networks, we need to consider the two vertex subsets of two-mode cores and define accordingly abstract closed bi-patterns. Each component of a bi-pattern is then associated to one mode. We also show that the same methodology applies to hub-authority cores of directed networks in which each vertex subset is associated to a role (in or out). We illustrate the methodology both on a two-mode network of epistemological data and on a directed advice network of lawyers.

Local rules associated to k-communities in an attributed graph

We address the problem of finding local patterns and local rules in an attributed graph. A (global) closed pattern is the most specific attribute pattern shared by the vertices of the (possibly simplified) subgraph induced by some attribute pattern. A local closed pattern is the maximal attribute pattern associated to a particular dense region of this subgraph. As such local regions, we are in particular interested in k-communities of pattern subgraphs. In this case we show that there is a closure operator such that, given a pattern q subgraph and a k-community in this subgraph, returns the local closed pattern shared by all the members of the community. We then consider how to generate triples (c, e, l) where c is a (global) closed pattern whose subgraph contains e as a k-community, and l is the corresponding local closed pattern. This leads to implication rules expressing what new attributes are specific of the k-community e in the pattern c subgraph.

Learning First Order Rules from Ambiguous Examples

We investigate here relational concept learning from examples when we only have a partial information regarding examples: each such example is qualified as ambiguous as we only know a set of its possible complete descriptions. A typical such situation arises in rule learning when truth values of some atoms are missing in the example description while we benefit from background knowledge. We first give a sample complexity result for learning from ambiguous examples, then we propose a framework for relational rule learning from ambiguous examples and describe the learning system LEAR. Finally we discuss various experiments in which we observe how LEAR copes with increasing degrees of incompleteness.

A Learning and Intentional Local Policy Decision Point for Dynamic QoS Provisioning

In the policy-based network management, the local policy decision point (LPDP), is used to reach a local decision. This partial decision and the original policy request are next sent to the PDP which renders a final decision. In this paper, we propose to give a real autonomy to the LPDP in term of internal decision and configuration. The LPDP is considered as a learning BDI agent that autonomously adapts the router’s behavior to environment changes