Séminaire de Recherche du CRI Novembre 2021

SAIaaS: A Blockchain-based solution for secure artificial intelligence as-a-Service
 Andrea Chretien (IKSEM 2020-2021)

Artificial Intelligence models are crucial elements to support many sectors in the current global economy. Training those models re- quires 3 main assets: data, machine learning algorithms, and processing capabilities. Given the growing concerns regarding data privacy, algo- rithm intellectual property, and server security, combining all 3 resources to build a model is challenging. In this paper, we propose a solution al- lowing providers to share their data and run their algorithms in secured cloud training environments. To provide trust for both clients and asset providers in the system, a blockchain is introduced to support the ne- gotiation, monitoring, and conclusion of model production. Through a preliminary evaluation, we validate the feasibility of the approach and present a road map to a more secure Artificial Intelligence as-a-service.

Télécharger le mémoire https://s.miage.dev/1xKTz

Ramona Elali (Doctorante CRI sous la supervision de Rebecca  Deneckere et Camille Salinesi)

An intention mining approach using ontology for contextual recommendations

Most modern Information Systems record user activity logs to extract knowledge and support new and higher quality services. Recommender systems (RS) in particular those based on collaborative filtering, are an example of such new services that have gained tremendous popularity. However, we know that standard activity logs are too poor to properly reflect user activity as they are influenced by many environmental factors (time, location, weather, profile, etc.), out of which a lot is not grasped in standard user activity logs. This makes it very difficult to mine users' intentions, which are key to analyze users' behavior reliably, and therefore key to quality recommendations. Several types of sources, such as sensors, external systems, outside actors, domain knowledge bases, or forecast systems could be used to obtain this information. The theory in this PhD is that the combination of multiple types of sources can extensively contribute to providing better insight into user activityt hrough context-rich intentional process mining and therefore deliver newer and higher quality recommendations. While existing research mainly focuses on log datasets; only a few contributions consider ontologies to gather multiple sources. We propose a novel approach that (a) combines different types of sources into ontologies (b) uses such ontologies for intentional process mining (c) exploits this intentional model for contextual recommendations.

Keywords: Intention Mining, Information System, Logs, Ontology, Context,