Neural networks are becoming more prevalent in an increasing number of safety-critical autonomy applications. It is important to ensure that these data-driven autonomous systems are safe given challenges associated with deployment in the real world. To address this issue, many formal approaches have been developed using tools such as barrier functions, reachability analysis, contraction theory, and temporal logics. VNNCS seeks to bring together academic experts in the field of neural system verification and industry stakeholders with the goal of identifying research gaps between what is currently possible and what is needed for practical application of neural system verification methods. This workshop will feature short submitted papers that highlight the latest innovations and provide opportunities for feedback on early research ideas, along with invited talks from leaders in the field and a panel discussion.

Submission Guidelines

Papers should be a total of 6 pages or less, not including references, in double column format. VNNCS is non-archival and may accept papers which are in review for other published venues, but not those which have already been published. Accepted papers will have the opportunity to showcase their work with a 10 minute presentation and/or during a poster session. Following is a list of suggested topics for submission, but related ideas are also encouraged:

• System verification
• Control barrier functions
• Reachability analysis
• Contraction theory
• Temporal logics
• Set propagation
• NN repair
• Out-of-distribution detection
• Conformal prediction

Organizing Committee

• Samuel Akinwande
• Michael Everett
• Sydney Katz
• Nicholas Rober
• Chelsea Sidrane
• Esen Yel

Contact

All questions about submissions should be emailed to Nick Rober [nrober@mit.edu]