AI Paper Atlas

Research map / Continual learning / Online and replay-based

COOM: A Game Benchmark for Continual Reinforcement Learning

Tristan Tomilin, Meng Fang, Yudi Zhang, Mykola Pechenizkiy

NeurIPS 2023 · Continual learning › Online and replay-based · approach: Plasticity-preserving continual reinforcement learning · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
make tasks harder (COC instead of CO8)PackNet · COOMaverage performance: 0.82 -> 0.20Hurt
make tasks harder (COC instead of CO8)MAS · COOMaverage performance: 0.58 -> 0.04Hurt
lengthen cross-domain sequence from 4 to 8 tasksMAS · COOM cross-domainaverage performance: 0.55 -> 0.82Improved
lengthen cross-domain sequence from 4 to 8 tasksL2 · COOM cross-domainaverage performance: 0.71 -> 0.87Improved
lengthen cross-objective sequence from 4 to 8 tasksL2 · COOM cross-objectiveaverage performance: 0.80 -> 0.71Hurt
lengthen cross-objective sequence from 4 to 8 tasksMAS · COOM cross-objectiveaverage performance: 0.72 -> 0.58Hurt
prepend first-half tasks (CO82nd vs CO4)all methods · COOM cross-objectiveaverage performance: 0.618 -> 0.643Improved
prepend first-half tasks (CD82nd vs CD4)all methods · COOM cross-domainaverage performance: 0.595 -> 0.673Improved
use AGEM instead of naive fine-tuningSAC · COOM averageaverage performance: 0.40 -> 0.28Hurt
use VCL instead of naive fine-tuningSAC · COOM averageaverage performance: 0.40 -> 0.33Hurt
use PackNet instead of fine-tuningSAC · COOM averageforgetting: 0.53 -> 0.01Improved
weaken regularization strength in L2/EWCL2 and EWC · COOMstability and plasticityNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy