Through the EPI project, SystemX and its partners aim to evaluate and optimize the performance of decision-making systems that embed artificial intelligence (AI) such as machine learning (neural networks), in the domains of autonomous transport as well as marine and other uncertain complex environments.
Bringing together four industrial partners (Apsys, Expleo, PSA Group and Naval Group) around the common interest of evaluating the performance of AI-based decision systems, the EPI (AI-based Decision Making Systems’ Performance Evaluation) project led by SystemX serves three purposes:
- Evaluating the performance of decision systems based on neural network algorithms (e.g. exact representation of reality, latency of the system, safety of operation, etc.).
- Improving the performance of neural network systems by mixing real and simulated data. Indeed, the numerical simulation makes it possible to increase the body of studied data and thus to increase the performances of the algorithms.
- Proposing an approach to evaluate the level of coverage of appropriate situations, based on two cases of applications. In the field of the autonomous vehicle, the studied function will be of the autopilot type. In the case of the maritime environment, the decision-making function remains to be defined but may concern port entries or exits, dense areas and multi-ships, etc.
The shared ambition of the project’s stakeholders is to build a methodology for evaluating the performance of decision systems based on machine learning (neural networks). This method must guarantee the independence between the (simulated) learning data and the test data (real data) and ensure the traceability and reproducibility of the results.
Two Proof of Concept(PoC) will be carried out in the fields of autonomous road and maritime transport.