# Colla competition - Sequoia This is an outline of the proposal for a new competition for a new Continual Learning conference (CoLLA). The competition would be centered around [Sequoia](https://www.github.com/lebrice/Sequoia), a framework built to facilitate research in both Continual Supervised Learning and Continual Reinforcement Learning). - This could possibly involve hiring an intern? ## High-Level Deliverables - Hosting solution - Need to have a way to run a user's method on the benchmarks of the competition (training + evaluation) - Previous Sequoia competition used a Docker-based hosting solution (managed by Florian Golemo) - Demo / cookiecutter repository - Existing repo is docker-based: https://github.com/fgolemo/cl-workshop-cvpr21-docker - Would it be better to use Singularity? - Sequoia updates - The Sequoia repo has already contains the benchmarks needed for the competition - (Optional): Improve documentation, examples, tests, etc. ## CoLLA - effort by MILA (irina, sarath) and McGill (Doina) to host the first ever conference on - Lifelong Learning / Continual Learning, - Meta-Learning - Multi-task Learning - etc - other parties are DeepMind (Pascanu), AMI (martha white), Google Brain (hugo larochelle) - there will be a lot of eyes on the first conference - good publicity for IDT - also, it should be in Montreal... ## The competition will be mostly a copy-pasta of [last year's competition](https://eval.ai/web/challenges/challenge-page/829/overview) - there will be 2 tracks, one supervised learning and one reinforcement learning one. - competitors are encourage attempt both tracks with the same solution. - we are pushing the field build methods that are neither SL nor RL specific - RL track = Monsterkong - main dep: meta_monsterkong - SL track = CtRL - main dep: ctrl - other important dep: - pytorch, pytorch lightning, others?