AI Paper Atlas

Research map / Reinforcement learning / Offline RL

Hokoff: Real Game Dataset from Honor of Kings and its Offline Reinforcement Learning Benchmarks

Yun Qu, Boyuan Wang, Jianzhun Shao, Yuhang Jiang, Chen Chen, Zhenbin Ye, Lin Liu, Yang Jun Feng, Lin Lai, Hongyang Qin, Minwen Deng, Juchao Zhuo and 6 more

NeurIPS 2023 · Reinforcement learning › Offline RL · approach: Offline RL sample complexity theory · 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
use QMIX+CQL instead of CQL (MARL structure for structured action space)CQL · HoK1v1 hard_expertwinning rate: 0.30±0.01 -> 0.44±0.05Improved
use QMIX+CQL instead of CQLCQL · HoK1v1 norm_generalwinning rate: 0.20±0.04 -> 0.32±0.03Improved
use TD3-based TD3+BC instead of BCBC · HoK1v1 norm_expertwinning rate: 0.64±0.01 -> 0.03±0.01Hurt
use OMAR instead of IND+CQLIND+CQL · HoK3v3 norm_expertwinning rate: 0.84±0.06 -> 0.39±0.16Hurt
add communication (COMM+CQL vs IND+CQL) (ablation)IND+CQL · HoK3v3 norm_mixedwinning rate: 0.46±0.12 -> 0.48±0.12Improved
add communication (COMM+CQL vs IND+CQL) (ablation)IND+CQL · HoK3v3 norm_multi_herowinning rate: 0.3±0.07 -> 0.31±0.2No clear effect
use CTDE (MAICQ) instead of independent learning (IND+ICQ) (ablation)IND+ICQ · HoK3v3 norm_expertwinning rate: 0.65±0.12 -> 0.61±0.09Hurt
use CTDE (MAICQ) instead of independent learning (IND+ICQ) (ablation)IND+ICQ · HoK3v3 hard_generalwinning rate: 0.31±0.09 -> 0.14±0.06Hurt
hard-level instead of norm-level datasetQMIX+CQL · HoK1v1 expertwinning rate: 0.67±0.01 -> 0.44±0.05Hurt
hard-level instead of norm-level datasetIND+CQL · HoK3v3 expertwinning rate: 0.84±0.06 -> 0.67±0.04Hurt
expert partner instead of stupid partnerIND+CQL · HoK3v3 heterogeneouswinning rate: 0.24±0.17 -> 0.57±0.04Improved
multi-task training instead of generalization datasetsQMIX+CQL · HoK1v1 normwinning rate: 0.32±0.03 -> 0.41±0.02Improved

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