Internship R&D at ENGIE LAB – Combining Bayesian Network and Ontology H/F

Job description

ENGIE is a global player in the energy sector, committed to the energy transition and expert in 3 business lines: electricity, natural gas, energy services.

ENGIE recruits thousands of innovative and motivated professionals around the world, to embody the future of the energy sector and be at the service of its customers. Interns will work closely with a group of researchers and engineers on fulfilling and innovative projects, promoting agility and creativity to meet the energy challenges of today and the future.

ENGIE Lab CRIGEN is the centre of research, development and operational expertise dedicated to gas, new energies and emerging technologies. Located in the Paris region in the city of Stains, it has about 200 employees. It supports the business by providing cutting edge expertise and develops tested, proven and marketable industrial applications. CRIGEN is committed to sharing novel ideas, scientific knowledge and technical expertise and its ability to innovate is a key advantage for the ENGIE Group.

Context:
An ontology is well known to be the best way to represent knowledge in a domain of discourse. It is defined by Gruber as “an explicit specification of a conceptualization”. It allows to represent explicitly and formally existing entities, their relationships and their constraints in an application domain. This representation is the most suitable and beneficial way to resolve many challenging problems related to information domain (e.g., semantic interoperability among systems, knowledge sharing, and knowledge capitalization). Ontology formalization can be based on First order logic (FOL) to describe concepts, relationships and constraints, enabling it to make inferences and giving it a graphical representation. Using ontology has many advantages, among them we can cite: ontology reusing, reasoning and explanation, commitment and agreement on a domain of discourse, ontology evolution and mapping, etc.

Over the last 30 years, another representation model called Bayesian Network has emerged as a practically feasible framework of expert knowledge encoding and as a new comprehensive data analysis framework.

Bayesian networks, also referred to as Belief networks have emerged as one of the most successful tool for diagnosis tasks and have been applied in many real domain applications (diagnosis, machine diagnosis, etc.). BNs offer mechanisms to accurately represent the dependences between random variables and to perform automated reasoning under uncertainty. They are supplied with fast inference engines that enable to answer efficiently various types of probabilistic queries (computation of marginal, a priori, a posteriori, probabilities, of most probable explanations, of maximum a posteriori, etc.).

In practice, the combination of Bayesian networks and ontologies might be beneficial to have high expressiveness and reasoning possibilities under uncertainty. Despite the difference between these two domain representation models, they have the potential to complement each other: part of the value of ontology baseline knowledge may be used to enhance BN by resolving challenging tasks: (i) the identification of relevant variables (variable selection), (ii) the determination of structural relationships between the considered variables (arcs), and (iii) the estimation of parameters which are represented by conditional probability tables (CPTs) associated to for each node in the BN model. Once the Bayesian network is learned, its results can be used together with ontology reasoning engine to perform probabilistic inference.

The internship aims to propose a new approach for combining Bayesian network and ontology in order to improve the analysis of the knowledge. The trainee will deeply study the state-of-the-art approaches based on recent and future advances in Semantic Bayesian Networks and Probabilistic Ontologies.

[1] Costa, P., Kathryn B. Laskey and Ghazi AlGhamdi. “Bayesian ontologies in AI systems.” (2006).

[2] Ben Ishak, Mouna & Leray, Philippe & Ben Amor, Nahla. (2011). A two-way approach for Probabilistic Graphical Models structure learning and ontology enrichment. KEOD 2011 - Proceedings of the International Conference on Knowledge Engineering and Ontology Development. 189-194.

[3] Emna, Hlel & Jamoussi, Salma & Ben Hamadou, Abdelmajid. (2014). Intégration d'un réseau bayésien dans une ontologie.

[4] Stefan Fenz, An ontology-based approach for constructing Bayesian networks, Data & Knowledge Engineering, Volume 73, 2012, Pages 73-88.

Trainee profile
M2, computer engineering school, you have a technical profile in Bayesian Network, knowledge on Semantic Web, and machine learning.

Requirements:
• Good knowledge (theoretical and applied) in Bayesian networks, Semantic Web, Machine Learning (ML) ;
• Strong knowledge skills on Python/R and ML frameworks; Self-driven and comfortable

Other details:

·  Duration: 6 months contract for start-up as soon as possible.
·  Please attach your CV and cover letter.
·  Localisation : CRIGEN (Centre de Recherche et d’Innovation dans le Gaz et les Energies Nouvelles) ENGIE in 4, rue Joséphine Baker 93240 - Stains - (RER D, Tram 11).
·  Contacts: Please send CV, cover letter, academic transcripts to contacts below, with Email Subject :[Internship R&D at ENGIE LAB – Automatic Summarization for Contention analysis]
·  Ahmed.mabrouk@engie.com
·  lynda.temal@engie.com
·  Philippe.calvez1@engie.com
·  sarra.ben-abbes@external.engie.c om

Additional Information
·  Posting Date: Dec 14, 2020