Research map / Reinforcement learning / Policy gradient and actor-critic
Informed Asymmetric Actor-Critic: Leveraging Privileged Signals Beyond Full-State Access
Daniel Ebi, Damien Ernst, Klemens Böhm, Gaspard Lambrechts
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use informed critic with signal [s1,s2,s4,s5] instead of symmetric history critic | informed-asym-A2C · Synthetic informed POMDP (|S|=20, 5 state features) | AUC of mean episodic return: 1.06 x 10^{5} -> 1.23e+05 | Improved |
| add reward-relevant features s4 and s5 to the privileged signal [s1,s2] (ablation) | informed-asym-A2C · Synthetic informed POMDP (|S|=20, 5 state features) | AUC of mean episodic return: 1.07e+05 -> 1.23e+05 | Improved |
| add s3 (redundant feature) to the signal [s1,s2,s4,s5] (ablation) | informed-asym-A2C · Synthetic informed POMDP (|S|=20, 5 state features) | AUC of mean episodic return: 1.23e+05 -> 1.19e+05 | Hurt |
| add s3 to the signal [s1,s2,s4,s5] (ablation) | informed-asym-A2C · Synthetic informed POMDP (|S|=20, 5 state features) | Prediction informativeness L_tau: 0.064 -> 0.056 | Hurt |
| add s3 to the signal [s1,s2,s4,s5] (ablation) | informed-asym-A2C · Synthetic informed POMDP (|S|=20, 5 state features) | Residual informativeness p-value: 0.035 -> 0.038 | Hurt |
| add s3 to the signal [s1,s2] (ablation) | informed-asym-A2C · Synthetic informed POMDP (|S|=20, 5 state features) | AUC of mean episodic return: 1.07e+05 -> 1.08e+05 | No clear effect |
| add s4 to the signal [s1,s2] (ablation) | informed-asym-A2C · Synthetic informed POMDP (|S|=20, 5 state features) | Prediction informativeness p-value: 0.801 -> 0.043 | Improved |
| add s5 to the signal [s1,s2] (ablation) | informed-asym-A2C · Synthetic informed POMDP (|S|=20, 5 state features) | AUC of mean episodic return: 1.07e+05 -> 1.10e+05 | No clear effect |
| add s4 to the signal [s1,s2,s3] (ablation) | informed-asym-A2C · Synthetic informed POMDP (|S|=20, 5 state features) | Residual informativeness p-value: 0.170 -> 0.068 | Improved |
| add s5 to the signal [s1,s2,s4] (ablation) | informed-asym-A2C · Synthetic informed POMDP (|S|=20, 5 state features) | AUC of mean episodic return: 1.08e+05 -> 1.23e+05 | Improved |
| use informed partial-signal critic instead of full-state history-state critic | informed-asym-A2C · POPGym Concentration | episodic return (Concentration) | Improved |
| use informed critic instead of symmetric A2C on Position Cart Pole | informed-asym-A2C · POPGym Position Cart Pole | episodic return (Position Cart Pole) | Improved |
Closest papers by meaning
- A Theoretical Justification for Asymmetric Actor-Critic Algorithms (ICML 2025 · similarity 0.91)
- Provable Partially Observable Reinforcement Learning with Privileged Information (NeurIPS 2024 · similarity 0.90)
- PIGDreamer: Privileged Information Guided World Models for Safe Partially Observable Reinforcement Learning (ICML 2025 · similarity 0.89)
- Studying the Interplay Between the Actor and Critic Representations in Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Dual Critic Reinforcement Learning under Partial Observability (NeurIPS 2024 · similarity 0.88)
- ReLU to the Rescue: Improve Your On-Policy Actor-Critic with Positive Advantages (ICML 2024 · similarity 0.88)
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.
