AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Procedure Planning in Instructional Videos via Contextual Modeling and Model-Based Policy Learning

Jing Bi, Jiebo Luo, Chenliang Xu

ICCV 2021 · Reinforcement learning › Visual and representation RL · approach: Inferring decision-making from behavior · 38 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
add accumulated-reward maximization over the whole trajectory (ablation)Int-MGAIL · CrossTask procedure planning (T=3)success rate: 7.18 -> 17.03Improved
add accumulated-reward maximization over the whole trajectory (ablation)Ext-MGAIL · CrossTask procedure planning (T=3)success rate: 15.18 -> 21.27Improved
add the discriminator (adversarial imitation) loss to supervised sequence learning (ablation)Int-MGAIL · CrossTask procedure planning (T=3)success rate: 5.89 -> 17.03Improved
add the discriminator (adversarial imitation) loss to supervised sequence learning (ablation)Ext-MGAIL · CrossTask procedure planning (T=3)success rate: 11.42 -> 21.27Improved
add hindsight relabeling of demonstrations (ablation)Int-MGAIL · CrossTask procedure planning (T=3)success rate: 14.39 -> 17.03Improved
add hindsight relabeling of demonstrations (ablation)Ext-MGAIL · CrossTask procedure planning (T=3)success rate: 18.01 -> 21.27Improved
make the transition model/policy stochastic (Ext vs Int)MGAIL · CrossTask procedure planning (T=3)success rate: 17.03 -> 21.27Improved
make the transition model/policy stochastic (Ext vs Int)MGAIL · CrossTask procedure planning (T=4)success rate: 9.47 -> 16.41Improved
remove HER from the stochastic model (Ext w/o HER versus Int-MGAIL)Ext-MGAIL · CrossTask procedure planning (T=3)success rate: 17.03 -> 18.01Hurt
replace DDN with model-based imitation learning (Int-MGAIL)Int-MGAIL · CrossTask procedure planning (T=3)success rate: 12.18 -> 17.03Improved
use Ext-MGAIL instead of Int-MGAIL on a second datasetMGAIL · second dataset (T=3)success rate: 20.19/22.11No clear effect
use Ext-MGAIL versus DDN for walk-through orderingExt-MGAIL · CrossTask walk-through planning (T=3)pair accuracy: 86.81 -> 93.66Improved

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