Jason Induction of Logical Decison Trees (JILDT) is a library that provides a learning mechanism based on induction of logical decision trees to implement learner agents in Jason, the well-known Java-Based implementation of AgentSpeak(L).

Top-down Induction of Logical Decision Trees is an Inductive Logic Programming technique, adopted for learning in the context of rational agents. The first-order representation of Tilde is adequate to form training examples as sets of beliefs, e.g., the beliefs of the agent supporting the adoption of a plan as an intention; and the obtained hypothesis are useful for updating the plans and beliefs of the agents, i.e., a Logical Decision Tree expresses hypotheses about the successful or failed executions of the intentions.

Agents defined as instances of this class are able to learn about their reasons to adopt intentions based on their own experience. Two levels in the inductive process have been implemented: A Java-based level with computational performance in mind; and an AgentSpeak(L)-based level which opens the door for some particular forms of social learning. A set of internal actions and plans are provided for allowing the agents to autonomously perform inductive experiments. Implementation details can be reviewed in the published papers.

More details can be found in


JaCaDDM is an agents & artifacts oriented Distributed Data Mining (DDM) tool that entails the creation and testing of Learning Strategies. A Learning Strategy is an encapsulated DDM workflow modeling the interaction of agents with their environment (which includes DM tools) and other agents, with the objective to create a classification model from distributed data.

Learning strategies are meant to be general in the sense that they can be applied to any distributed setting, deployment details are managed by the JaCaDDM platform. In this sense, learning strategies are plug-and-play. The JaCaDDM distribution already comes with a set of different learning strategies to try, and it is also possible to add new ones.

JaCaDDM considers as a distributed setting any kind of environment where data is split in various sites (even geographically distributed). With JaCaDDM is possible to configure and launch a deployment that takes into account the different sites, and their data, that participate in the DDM process. As mentioned, the actual process is encapsulated on the learning strategy, which may have some configurable parameters that can be set as part of the general configuration.

JaCaDDM provides a tool to experiment and do research with different DDM approaches, as it makes an evaluation of the produced classification model, yielding various performance statistics (total time, classification accuracy, network traffic produced, model complexity, confusion matrix).

JaCaDDM can be extended through the adding of new learning strategies and artifacts. Artifacts are first-class entities in the agent environment that encapsulate services, in the case of JaCaDDM, these services consists on DM related tools.

More details can be found in

Publicado en Páginas