AI Paper Atlas

Research map / Reinforcement learning / Imitation learning

Invariant Causal Imitation Learning for Generalizable Policies

Ioana Bica, Daniel Jarrett, Mihaela van der Schaar

NeurIPS 2021 · Reinforcement learning › Imitation learning · approach: Behavior cloning theory and models

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
ICIL instead of BCImitation policy · MIMIC-III mechanical ventilatorACC: 0.783 ± 0.001 -> 0.855 ± 0.003Improved
ICIL instead of VDICE, the best baselineImitation policy · MIMIC-III mechanical ventilatorAUC: 0.784 ± 0.001 -> 0.856 ± 0.004Improved
ICIL instead of VDICE, the best baselineImitation policy · MIMIC-III mechanical ventilatorAPR: 0.716 ± 0.001 -> 0.789 ± 0.004Improved
ICIL instead of EDMImitation policy · MIMIC-III mechanical ventilatorAUC: 0.741 ± 0.011 -> 0.856 ± 0.004Improved
Add IRMv1 penalty to BCBC · MIMIC-III mechanical ventilatorACC: 0.783 ± 0.001 -> 0.791 ± 0.002No clear effect
Add IRMv1 penalty to RCALRCAL · MIMIC-III mechanical ventilatorAUC: 0.771 ± 0.002 -> 0.766 ± 0.003No clear effect
Add IRMv1 penalty to VDICEVDICE · MIMIC-III mechanical ventilatorACC: 0.794 ± 0.001 -> 0.766 ± 0.001Hurt
Add IRMv1 penalty to VDICEVDICE · MIMIC-III mechanical ventilatorAUC: 0.784 ± 0.001 -> 0.730 ± 0.001Hurt
Add IRMv1 penalty to EDMEDM · MIMIC-III mechanical ventilatorAUC: 0.741 ± 0.011 -> 0.717 ± 0.015Hurt
Add IRMv1 penalty to EDMEDM · MIMIC-III mechanical ventilatorAPR: 0.682 ± 0.005 -> 0.673 ± 0.007Hurt
Add IRMv1 invariance penalty to existing batch imitation methodsBC, RCAL, VDICE, EDM · OpenAI gym tasks (Acrobot, Cartpole, LunarLander, BeamRider)average returnNo clear effect
ICIL instead of all benchmarksImitation policy · OpenAI gym tasks (Acrobot, Cartpole, LunarLander, BeamRider)average return on unseen test environmentImproved

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