Here are the latest news on CAISAR:
We are proud to announce that our analyzer PyRAT scored second (out of eight participants) during the Fifth International Verification of Neural Networks Competitions (VNN-COMP 2024). VNN-COMP was co-hosted with the conference Computer Aided Verification (CAV) and the Symposium on AI Verification (SAIV).
PyRAT verified more than 1700 properties on multiple benchmarks, ranging from classical convolutional neural networks to GAN architectures and cyberphysical systems. We will update this page…
On the occasion of the 34th birthday of the abolition of apartheid laws in South Africa, we are honoured to release CAISAR version 2.0.
The release source is available at our public forge. As our last releases, CAISAR will soon be available on opam and on Dockerhub.
Here are some of the most prominent features we added.
Specification and verification of several neural networks at once
CAISAR specification language already allowed to write specifications that involved several neural networ…
For its first participation, during the Fourth International Verification of Neural Networks Competition (VNN-COMP 2023), the abstract analyzer PyRAT finished at the third place by verifying more than 400 properties on 8 different benchmarks. Congratulations to all competitors! For more details, you may consult the full report of the competition.
(all right reserved to jameschipper / Adobe Stock images)
On the occasion of the 68th birthday of the Flag of Europe, we are delighted to release the version 1.0 of CAISAR. The release source is available under our gitlab. It will soon be available on opam and on Dockerhub.
Extensions of specification language and interpretation capabilities We extended the WhyML specification language to take into account common machine learning constructs. It is now possible to model machine learning computations using vectors and datasets. Parts of this specific…
Last week, we presented CAISAR at the workshop on Safe and Robust Machine Learning.
Among the workshop organizers were member of the Lab for AI Verification. The LAIV and CAISAR teams have close research interests, and it was a pleasure to discuss with the diverse panel of participants on the topic of verification of machine learning.
The discussions provided enlightening insights on the needs in both the industry and research.
We are happy to announce the release of CAISAR 0.2, to celebrate the 234th birthday of the Serment du jeu de Paume. On this day, courageous people took a vow to abolish autocracy and fight for their rights. This release is the result of more than one year of work. Here are some of the most prominent changes that we are happy to share. The release is available under our gitlab, on opam and on Dockerhub.
New prover support We added the support of several provers:
the alpha-beta-CROWN prove…
The JFLA (French for Journées Francophones des Langages Applicatifs) is a yearly meeting of both academics and industrials that share a common interest on functional programming - and incidently, software verification.
Industry side, companies OCamlpro, Mitshubishi Electric, TrustInSoft were present. From the academia, people from Formal Methods Laboratory, ENSIIE and INRIA were attending.
We gave a 15-minute presentation of CAISAR, available here (in French). The relevant publication is av…
We are delighted to release CAISAR’s technical report, available on HAL.