Flavio Corradini

Independent organizations executing distributed processes must jointly satisfy regulations, contractual obligations, and industrial standards. Verifying this demands evidence of the whole behavior, which is fragmented across their information systems. Case-centric event logs flatten the many-to-many relations among the entities involved, while object-centric techniques preserve them, and lack cross-organizational-specific solutions.We extend the object-centric event data meta-model with an object sharing procedure based on lazy replication, connecting the event logs of different organizations. On this foundation we define a language of object-centric compliance rules combining log filters, which select events through conditions on event payloads, object attributes and event-to-object relationships, with behavioral constraints correlating events along lineages of object-to-object relationships under control-flow patterns and temporal bounds. Evaluation returns the matching and the violating executions, identifying the events and objects accountable for each result.The language is demonstrated through a prototype checker applied to two distributed logs, order management and container logistics, and its semantics receives a declarative counterpart through a translation into first-order logic, which reduces compliance checking to query answering over a finite structure.

Renal insufficiency is a frequent and serious side effect of cancer and frequently is compounded by cytotoxic chemotherapy and nephrotoxic medications. Conventional approaches, such as serum creatinine and eGFR measurement have low sensitivities to detect early changes. Here, we present an Artificial Intelligence (AI) driven Clinical Decision Support System (CDSS) focused on estimating kidney dysfunction risk based on real-world cancer care data. With supervised machine learning such as Random Forest, XGBoost, Support Vector Machines, and Multilayer Perceptrons, we examined the discriminative model for 5,000 patient cases. Preprocessing of raw data included handling the missing values, feature selection, calibrating classes with the help of SMOTE. No model had better performance than XGBoost, whose F1-score was 0.82 and AUC was 0.88. We generated global explanations identifying the most influential features (e.g., eGFR, creatinine, tumor type) using SHAP summary plots, and local patient level explanations illustrating individual risk contributions using LIME and SHAP waterfall plots. Our experimental results demonstrated that AI algorithms can be used in a CDSS to aid with early intervention, improve kidney safety, and personalize the treatment plans for oncology. While the data were single center in origin and there was not external validation, the system demonstrates significant potential for integration into electronic medical records and testing across different clinical backgrounds is needed.

While decentralized applications adoption has surged in several domains, the smart contracts implementing them can diverge from their specified logic. To address this issue, the data generated during the execution of smart contracts can provide valuable insights into the actual behavior of an application, enabling its analysis. For this reason, we employ compliance checking to audit decentralized applications based on their execution data. To enable such a technique, we provide a threefold contribution. Firstly, we present a domain-specific language, called CoBlock, to define compliance rules by relying on blockchain characterizations as a first-class citizen. Then we designed a visual rule modeling solution, CoBlockly, which abstract the underlying language, to facilitate auditors on building blockchain-specific compliance rules. Finally, we implemented a tailored framework for blockchain compliance checking to streamline the process of defining and checking compliance rules over decentralized applications execution data. The evaluation, which employs the real-world CryptoKitties decentralized application, includes accuracy and performance analysis in addition to a usability assessment to measure the feasibility and to demonstrate the practicality of the framework.