On October 2nd, 2024, our research group presented its latest work on combining Artificial Intelligence (AI) and Operations Research (OR) at Google. ORAILIX is dedicated to developing efficient, reliable, and understandable models that address real-world challenge, through integrating structured knowledge and business skills into our approaches. We aim to provide optimal, safe, and explainable solutions that minimize resource use and vulnerabilities.

A core aspect of our research is the development of responsible and trustworthy AI systems. We strive to reduce the environmental impact of AI by controlling model sizes and energy consumption. Additionally, we tackle security challenges such as data leakage and model vulnerabilities. Our goal is to create AI systems that facilitate fair and ethical decision-making while enhancing model transparency, robustness, and traceability. Our work is validated through collaborations with industry partners like Crédit Agricole, SNCF, Safran, Renault, and Orange, focusing on areas such as AI trustworthiness, mobility optimization, predictive maintenance, and cloud architecture enhancement.

During the presentation, we highlighted a pipeline we developed to identify and address memory-related issues in large language models. This solution contributes to the creation of more secure and reliable AI systems and has been tested in various empirical settings. Our approach to modeling, efficiency, and frugality can be summarized as follows:

  • Modelling

    We emphasize the efficient formulation of problems to deliver reliable, safe, explainable, and optimal solutions.

  • Efficiency

    Our methods integrate structured knowledge and business skills directly into models, enhancing their practical relevance and performance.

  • Frugality

    We aim to reduce the size of models and datasets, minimizing system resources and vulnerabilities, and ensuring that our solutions are both effective and sustainable.