Data analytics has an ever increasing impact on tackling various societal challenges. In this article, we investigate how data from several heterogeneous online sources can be used to discover insights and make predictions about the spatial distribution of crime in large urban environments. A series of important research questions is addressed, following a purely data-driven approach and methodology. First, we examine how useful different types of data are for the task of crime levels prediction, focusing especially on how prediction accuracy can be improved by combining data from multiple information sources. To that end, we not only investigate prediction accuracy across all individual areas studied, but also examine how these predictions affect the accuracy of identified crime hotspots. Then, we look into individual features, aiming to identify and quantify the most important factors. Finally, we drill down to different crime types, elaborating on how the prediction accuracy and the importance of individual features vary across them. Our analysis involves six different datasets, from which more than 3,000 features are extracted, filtered, and used to learn models for predicting crime rates across 14 different crime categories. Our results indicate that combining data from multiple information sources can significantly improve prediction accuracy. They also highlight which features affect prediction accuracy the most, as well as for which particular crime categories the predictions are more accurate.

Abstract. This paper presents an argumentation mechanism for reconciling conflicts between planning agents related to plan proposals, which are caused by inconsistencies between basic beliefs regarding the state of the world or the specification of the planning operators. We introduce simple and efficient argument moves that enable discussion about planning steps, and show how these can be integrated into an existing protocol for belief argumentation. The resulting protocol is provably sound with regard to the defeasible semantics of the resulting agreements. We show how argument generation can be treated, for the specific task of argumentation about plans, by replacing the burden of finding proofs in a knowledge base by guided search.

No abstract

This paper proposes the β-WoLF algorithm for multiagent reinforcement learning (MARL) in the stochastic games framework that uses an additional "advice" signal to inform agents about mutually beneficial forms of behaviour. β-WoLF is an extension of the WoLF-PHC algorithm that allows agents to assess whether the advice obtained through this additional reward signal is (i) useful for the learning agent itself and (ii) currently being followed by other agents in the system. With this, agents are able to decide autonomously whether to follow the advice or not, safeguarding themselves against malicious or unreliable advice which, if followed, might lead them to sacrifice their own future rewards, as well as unilateral cooperation that could be exploited by other agents in the system. We report on experimental results obtained with this novel algorithm which indicate that it enables cooperation in scenarios in which the need to defend oneself against exploitation results in poor coordination using existing MARL algorithms. We present a critical discussion of its merits and limitations, and discuss its significance as a step toward the development of MARL algorithms capable of dealing with more complex forms of potentially unreliable communication.

No abstract