Call for Participation

Submission deadline: Extended to 22nd November 2022 Anywhere on Earth (AOE) 18th November 2022, Anywhere On Earth (AoE) . Submit on OpenReview

Notifications: 2nd December 2022

Bridge: 7-8th February 2023

The fields of causality and continual learning research complementary aspects of human cognition and are fundamental components of artificial intelligence if it is to reason and generalize in complex environments. On the one hand, causality theory provides the language, algorithms, and tools to discover and infer cause-and-effect relationships from data and a partial understanding of a complex system. On the other hand, continual learning systems learn over time from a continuous stream of data, enable knowledge transfer, and alleviate potentially malicious interference when distributional shifts are experienced. However, despite some recent interest in bringing the two fields together, it is currently unclear how causal models may describe continuous streams of data and, vice versa, for continual learning to exploit learned causal structure.

We propose to bring together the fields of continual learning and causality in a two-day AAAI-23 bridge program called “Continual Causality”. In this bridge program, we aim to take the first steps towards a unified treatment of these fields and to provide the space for learning, discussions, and to connect and build a diverse long-term community.

To this end, the Continual Causality Bridge invites submissions that present general positions and visions of how to link the two fields, outline challenges that need to be overcome, highlight synergies, and propose tangible future steps. Submissions of position papers should be up to two pages (excluding references, and without appendices) in the AAAI format. The submission deadline is November 22nd, 2022 (AOE). All submissions will be managed through OpenReview and will later be collected in a proceedings volume:

The review process is double-blind, so submissions should be anonymized. As our bridge’s focus is on building a long-term community, the review process will be light and inclusive, focusing primarily on technical correctness as a means of evaluation.

Our vision is for prospective community members to voice diverse views that have the potential to advance AI through an ongoing cross-disciplinary exchange. We therefore have no strict constraints on the exact sub-topics of submissions, as long as they target the overall goal of bridging the fields. Submissions are free to focus on a single particularly interesting synergy, take an angle from a specific scientific discipline, or sketch a grander view. To provide some suggestions, contributions could center around the following aspects, including but not limited to:

  • Continual learning and exploitation of causal systems in dynamic non-stationary environments.
  • Catastrophic interference and knowledge transfer in learning causal models in the context of continuous streams of data.
  • Effective ways for causal structure to aid in leveraging the accumulated knowledge of a continual learning system.
  • Leveraging causal tools to interpret distributional shifts in continual learning.
  • Next generation benchmarks that go beyond repurposing of existing datasets to adequately support the above items and further essential research questions towards a symbiosis of continual learning and causality.

For questions, join our Slack or Email the organizers at info[at]continualcausality[dot]org