Proceedings of the 2nd Conference on Defense Artificial Intelligence (CAID 2020) published on HAL

Proceedings of the 2nd Conference on Defense Artificial Intelligence (CAID 2020) published on HAL


The Defense Artificial Intelligence Conference, which took place remotely (due to the health situation) in December 2020, and has been organized since 2019 by Direction Générale de l’Armement (DGA) at the same time as the European Cyber ​​Week and the C & ESAR Conference, covering cyber security. The aim of this conference is to bring together defense players who publish scholarly articles on defense-related issues that include artificial intelligence. So there are several types of actors: institutions dedicated to defense such as the DGA itself or ANSSI, public institutions such as ONERA and CEA, schools and universities, such as Centrale Supélec and the University of Rennes, companies ranging from start-ups such as NukkAI to large groups such as Orange or Thales. The topics are also very different: the presentations are grouped into sessions covering cybersecurity, decision making, speed cameras, data security, and crisis management.

The aim of the conference is to bring together representatives who, although working in different fields, often encounter the same problems: reliability on the one hand – the consequences of error can be particularly heavy in this area – but also the ability to include these technologies, with all their It means in terms of reducing power consumption, or even respecting the confidentiality of data, which is particularly sensitive in this area. So themes emerge naturally, particularly notions of interpretability or trustworthy artificial intelligence. The conference allows people working in completely different worlds to meet and see which direction the state of the art is advancing.

Regarding the methods presented, there are a number of approaches that are widely used elsewhere: machine learning and deep learning, is widely used on many issues ranging from intrusion detection to radar target classification. These methods have also been the subject of research to reduce the cost associated with them so that you can simplify them, in particular by simplifying them. But beyond these methods, there are many references to so-called “hybrid” methods, including machine learning and so-called “symbolic” artificial intelligence, which is sometimes called Ai good old fashion or GOFAI. These methods generally include methods of learning based on semantic data, or even methods of reinforcement learning, which make it possible to ensure that the output will respect a certain number of properties with respect to the input. Sometimes relying on these two areas, a large number of articles—particularly those dealing with intrusion detection in cybersecurity—have focused on unsupervised methods of anomaly detection.

The organization of the conference was disrupted due to health conditions. The DGA has attempted to keep the conference face-to-face, which especially allows players on the field to meet and discuss potential opportunities. She finally had to resign to hold the teleconference. As such, I suggested a mixed process: authors and authors were invited to submit a video to avoid issues with live broadcasts, while still keeping time for questions from the audience. A few months after the conference, the platform on whichEuropean Internet Week He uploaded C & ESAR and CAID videos for a limited time, including videos submitted by authors and Q&A sessions. Proceedings of the conference were published on HAL at the end of April, while the call for papers for the 2021 edition will soon appear on the website of theEuropean Internet Week.


Source link