Intelligenza Artificiale

 Thinking to oneself is a prerogative of man when he needs to think about or repeat what he is doing or experiencing. It is a way of processing information and setting in motion a decision-making process. When this is done aloud, there is also a chance that someone else will understand the meaning or reasons for the action. Equipping an agent with the ability to reveal the reasons for its decisions is both a way to improve human interaction and a way to improve the triggering of a decision process. In this work, we propose to use the speech act to enable a coalition of agents to exhibit inner speech capabilities to explain their behavior, but also to guide and reinforce the creation of an inner model. The BDI agent paradigm, Jason, and CArtAgO are used to give agents the ability to act in a human-like manner. The BDI reasoning cycle has been extended to include inner speech. The proposed solution continues the research path that started with the definition of a cognitive model and architecture for human-robot teaming interaction and aims to integrate the believable interaction paradigm in it.

 This paper shows the capabilities offered by an integrated neural-logic multi-agent system (MAS). Our case study encompasses logical agents and a deep learning (DL) component, to devise a system specialised in monitoring flood events for civil protection purposes. More precisely, we describe a prototypical framework consisting of a set of intelligent agents, which perform various tasks and communicate with each other to efficiently generate alerts during flood crisis events. Alerts are only delivered when at least two separates sources agree on an event on the same zone, i.e. aerial images and severe weather reports. Images are segmented by a neural network trained over eight classes of topographical entities. The resulting mask is analysed by a Logic Image Descriptor (LID) which then submit the perception to a logical agent.

 Software agents are normally expected to operate in open and dynamic environments, and therefore they are often supposed to face situations that significantly deviate from the nominal course of events. The effective management of exceptional situations is of paramount importance to provide agents with the needed means to operate in their environments, mostly because these situations should be considered as the norm in open and dynamic environments. This paper presents some recent additions to the Jadescript agent-oriented programming language that were specifically designed to provide agents with the needed capabilities to effectively detect and manage exceptional situations. The first part of this paper motivates the need of sophisticated exception handling capabilities, also by relating the proposed language features with the state of the art documented in the literature. Then, the second part of this paper discusses the proposed language features, also considering the conceptual similarities and differences with the related features normally available in mainstream programming languages. In particular, the proposed language features are presented in terms of three language improvements: the general-purpose support to handle exceptions, the specific support to handle behaviour failures, and the specific support to handle stale messages. Finally, before concluding with some indications on future research activities, the third part of this paper describes a concrete example intended to practically present the actual use of the new language features.

 Semantic representation is a key enabler for several application domains, and the multi-agent systems realm makes no exception. Among the methods for semantically representing agents, one has been essentially achieved by taking a behaviouristic vision, through which one can describe how they operate and engage with their peers. The approach essentially aims at defining the operational capabilities of agents through the mental states related with the achievement of tasks. The OASIS ontology — An Ontology for Agent, Systems, and Integration of Services, presented in 2019 — pursues the behaviouristic approach to deliver a semantic representation system and a communication protocol for agents and their commitments. This paper reports on the main modelling choices concerning the representation of agents in OASIS 2, the latest major upgrade of OASIS, and the achievement reached by the ontology since it was first introduced, in particular in the context of ontologies for blockchains.

The XAI community is currently studying and developing symbolic knowledge-extraction (SKE) algorithms as a means to produce human-intelligible explanations for black-box machine learning predictors, so as to achieve believability in human-machine interaction. However, many extraction procedures exist in the literature, and choosing the most adequate one is increasingly cumbersome, as novel methods keep on emerging. Challenges arise from the fact that SKE algorithms are commonly defined based on theoretical assumptions that typically hinder practical applicability.

This paper focuses on hypercube-based SKE methods, a quite general class of extraction techniques mostly devoted to regression-specific tasks. We first show that hypercube-based methods are flexible enough to support classification problems as well, then we propose a general model for them, and discuss how they support SKE on datasets, predictors, or learning tasks of any sort. Empirical examples are reported as well –based upon the PSyKE framework –, showing the applicability of hypercube-based methods to actual classification tasks.

 We consider an extended version of sabotage games played over Attack Graphs. Such games are two-player zero-sum reachability games between an Attacker and a Defender. This latter player can erase particular subsets of edges of the Attack Graph. To reason about such games we introduce a variant of Sabotage Modal Logic (that we call Subset Sabotage Modal Logic) in which one modality quantifies over non-empty subset of edges. We show that we can characterize the existence of winning Attacker strategies by formulas of Subset Sabotage Modal Logic.

To the aim of constructing effective human-AI teams, that can be useful for improving caregiving in medicine and enhancing human performance also in other sectors (i.e., teaching), agents which interact with humans should be endowed with an emotion recognition and management module, capable of empathy, and of modeling aspects of the Theory of Mind, in the sense of being able to reconstruct what someone is thinking or feeling. In this paper, we propose an architecture for such a module, based upon an enhanced notion of Behavior Trees. We illustrate the effectiveness of the proposed architecture on a significant example and on a wider case study.

A cognitive-based routing algorithm is proposed. Concepts like local form and path algorithms are developed. Unlike current mainstream routing algorithms assume that all people know everything about the environment, the proposed algorithm allows people to have a complete or incomplete map knowledge and built up their own map knowledge in a piecemeal fashion. Using a hospital floor plan as the scenario, numerical experiments are conducted by assuming pedestrians to have different levels of map knowledge. Results show that reasonable routes could be frequently found even if pedestrians only have an incomplete knowledge of the network. Also pedestrians generally need to traverse more rooms if having zero or less map knowledge. Hence the proposed algorithm’s effectiveness is validated to some extent.

Click fraud is the sort of deception in which traffic figures for online ads are intentionally inflated. For businesses that advertise online, click fraud may occur often, resulting in erroneous click statistics and lost funds. That is why many businesses are hesitant to advertise their products on websites and mobile apps. To market their products safely, businesses need a reliable technique for detecting click fraud. In this paper we present a stacking algorithm as a solution to this problem. The proposed method’s premise is to combine multiple learners to achieve an optimal result. The Synthetic Minority Oversampling Technique (SMOTE) with a combination of undersampling are chosen to handle the unbalanced dataset. In the first-level learners, there are four supervised Machine Learning algorithms, which are AdaBoost, Random Forest, Decision Tree and Logistic Regression. Moreover, Logistic Regression is used again as a the second-level learner. To verify the efficacy of the suggested approach, comparative tests are carried out on the public dataset available on Kaggle from China’s largest independent big data service platform TalkingData. Multiple indicators, such as Accuracy, F1 Score, ROC curve, Loss Log and AUC Score, are utilized to analyze the prediction outcomes. The findings reveal that the stacking method improves forecast accuracy while also maintaining a high level of stability.

Representing uncertain information is crucial for modeling real world domains. This has been fully recognized both in the field of Logic Programming and of Description Logics (DLs), with the introduction of probabilistic logic languages and various probabilistic extensions of DLs respectively. Several works have considered the distribution semantics as the underlying semantics of Probabilistic Logic Programming (PLP) languages and probabilistic DLs (PDLs), and have then targeted the problem of reasoning and learning in them. This paper is a survey of inference, parameter and structure learning algorithms for PLP languages and PDLs based on the distribution semantics. A few of these algorithms are also available as web applications.