Affordable Industrial Artificial Intelligence

"The ""Affordable AI"" (AI2) proposal addresses three challenges blocking the AI revolution in industry: i) the design of smart systems requires considerable expertise and computational resources (AutoML); ii) ""black-box"" models do not necessarily meet the objectives of robustness, transparency and fairness; iii) in many industrial sectors, only ""small data"" are available. The AI2 consortium, composed of two world-class universities in AI (Univ. Paris-Saclay and University of Bonn), an SME specialized in AI (My-DataModels-MDM) and an SME specialized in the design and manufacture of pumps (Wilo), has all the skills required to address the above-mentioned issues, as follows: 1.AutoML will be extended to the multi-criteria optimization framework, in order to control /manage the compromise between performance and resources. 2.The black box models will be provided with a sensitivity analysis and a causal analysis. 3.We will use the human expert in the learning loop, to compensate for the lack of data, and the potentially incomplete specifications of the targeted smart system. The validation and transfer of the proposed functionalities will be carried out in three phases. Phase 1: The industrial partners provide the consortium with application cases (real or simulated data). Phase 2: The functionalities are developed and validated on these application cases. Phase 3: The functionalities will be adapted and evaluated in the context of similar and confidential industrial applications. The collaboration between industrial partners and laboratories will enable to consolidate and guarantee the robustness and generality of the technological advances made."