In addition to our call for papers at our third bridge on Continual Causality at AAAI-25 we are launching the Continual Confounding: ConCon Challenge!
In the ConCon challenge participants can submit contributions to solve one of two (or both) challenging scenarios surrounding the newly found phenomenon of continual confounding. In continual confounding, the difficulty extends beyond the traditional continual learning problem of catastrophic forgetting. Instead, task performance additionally deteriorates because each task is uniquely confounded! That is, even with a full memory of all past data, the sequential nature of confounders in ConCon poses a significant challenge – where joint training on all data succeeds, sequential training on accumulated tasks fails without further action!
Submission can be made until January 7th 2025 and all details can be found on the ConCon challenge’s Kaggle website. The most performant and innovative contributions will be invited to present their solutions at the AAAI bridge, whereas two highlighted contenders will further be awarded with a free bridge/lab/tutorial AAAI-25 registration.