Aim of the workshop:

As industry moves towards more inclusion of Artificial Intelligence (AI) in its products, there is a crucial need to investigate how academic and industrial research can contribute to the robustness assessment of AI in safety-critical systems. The present workshop aims at updating the community on the state-of-the-art in “Robustness in AI”, discussing how it aligns with industrial challenges and exploring a more collaborative future on AI safety between research institutes and industries.

This event gathers major actors in empirical (including Adversarial Machine-Learning) and formal methods for the robustness assessment of neural networks along with key industrial contributors.

It will specifically focus on:

• Empirical approaches to AI robustness assessment e.g. against unexpected / adversarial perturbations

• Formal / probabilistic methods to AI robustness assessment

• Compatibility, adaptability and scalability to industrial problematics of current techniques

• Potential limitations and ways forward towards their use in safety-critical systems.

Workshop format:

The workshop was initially intended as a week-long face-to-face meeting organised at the Lorentz Center in Leiden, near Amsterdam, NL (https://www.lorentzcenter.nl/). Due to the current pandemic and travel and sanitary restrictions, the workshop has now been converted to an online 4-day event. The workshop requires a commitment of 4hr/day adapted to our participants' availability and time zones. Participants may decide to only attend a fraction of the workshop, although they are strongly encouraged to attend the whole event as their inputs will be valued tremendously during cross-domain discussions.

The workshop programme can be found here.

The workshop will be split into Q&A sessions sparked by a series of pre-recorded reviews and technical lectures, group break-out sessions and panel discussions. Details on the break-out sessions can be found here.

Links to pre-recorded lectures are provided here.

Participants:

The majority of invitees includes researchers from institutes across the globe with expertise in AI robustness. We have also invited AI practitioners in industries of domains such as aviation, automotive, healthcare, or judicial where safety is of paramount importance. They will contribute to the discussions by presenting the current problems and challenges of the inclusion of AI in real-life applications. See participant list here.