ABSTRACT Accurately assessing the energy consumption associated with public buildings and the services they provide is essential for supporting sustainable infrastructure planning. This work presents an integrated modelling framework that combines machine learning prediction of building energy use with a graph based representation of user mobility, allowing a comprehensive estimation of total energy demand. The approach includes building operations, service related energy, travel by users and staff and supply chain logistics. Building energy consumption is predicted through supervised models trained on a curated subset of commercial facilities selected for their similarity to public service environments. Mobility is modelled using a synthetic geographical network that encodes population distribution, available transportation modes, behavioural tendencies and relocation dynamics. The framework is applied to a university reorganization scenario, exploring alternative facility configurations and varying degrees of remote activity. Results indicate that user travel is generally the dominant contributor to total energy demand, while the importance of building energy increases as in person attendance decreases. Sensitivity analyses confirm the robustness of the optimal configurations under different behavioural assumptions. We stress that this is not actually an optimization algorithm, but a parameter sweep; in real situations, constraints may be applied, for instance for building availability, costs, opportunity of sharing services or specific goals. The methodology is further demonstrated in a real healthcare application, where the predictive model enables reliable estimation of building energy use in the absence of direct measurements. Overall, the proposed framework illustrates how data driven modelling and intelligent system techniques can support sustainable decision making for complex public service infrastructures.

Natural Language Processing (NLP) algorithms have significantly advanced the capabilities of understanding, processing and generating human language. However, one persistent challenge in NLP is the problem of uncertainty, due e.g. to the inherent complexity of human language, variations in language usage across different contexts and domains, and the presence of noisy or incomplete data. In this work we take in consideration this problem for statistics derived from court documents by NLP systems.

Anonymization is an important aspect of data privacy protection, especially in the context of sensitive personal information collected through sensors. In this paper, we propose a new service-based architecture for anonymizing such data in real-time, ensuring that data is accessible to authorized users while maintaining privacy. Our architecture is based on the annotation of data at ingestion time, where privacy levels are assigned to sets of columns. The anonymization procedure is performed by compressing and encoding the data through an autoencoder model, where the encoder and decoder functions are defined as parametric functions composed of multiple hidden layers.

The use of data acquisition and fusion techniques allow to generate event graphs in support of criminal investigations. In this paper, an anonymized real case study will be presented to identify undue transactions through graph analysis. All the steps of an investigation protocol are illustrated by describing how the tools adopted in this paper allow to semi-automatically analyze huge amounts of data coming from different nature, identifying suspicious transactions with high precision.

Abstract Criminal investigation adopts Artificial Intelligence to enhance the volume of the facts that can be investigated and documented in trials. However, the abstract reasoning implied in legal justification and argumentation requests to adopt solutions providing high precision, low generalization error, and retrospective transparency. Three requirements that hardly coexist in today’s Artificial Intelligence solutions. In a controlled experiment, we then investigated the use of graph embeddings procedures to retrieve potential criminal actions based on patterns defined in enquiry protocols. We observed that a significant level of accuracy can be achieved but different graph reformation procedures imply different levels of precision, generalization, and transparency.

Events happening at defined moments in time and involving specific entities from a social or physical system can be organized in networks or graphs. The study of such event graphs may reveal causal relations between subsequent events or compound events that we define as “typed events”. Moreover, characteristic sequences of events or patterns can arise in consequence of phenomena affecting the system. Methods to build the event graph and to search for the typed events and their significance are described in detail. An embedding strategy to encode typed events in low dimensional vectors is defined, and both supervised and unsupervised learning is applied to search for meaningful patterns. Experiments have been conducted using data from a real investigation and some synthetic data.

Knowledge graphs are exploited in criminal investigation to integrate heterogeneous data sources and scale up the operational efficiency of enquiry protocols. Using a declarative perspective, protocols can be viewed as a set of data ingestion procedures and nested exact queries. This meets the probating nature of procedural justice that has to proceed from established facts. At the same time, the exact specification of queries represents a limit for enquiry protocols that can exclusively retrieve those facts in adherence to the designed queries. We then investigated the use of graph em-beddings procedures to extend the scope of a protocol by returning sub-graphs partially matching to its specification. Because exploring the entire set of sub-graphs quickly become computationally intractable, we developed an approach based on a hierarchical filtering procedure. A controlled experiment we executed has shown the feasibility of our approach.