AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

Model-Value Inconsistency as a Signal for Epistemic Uncertainty

Angelos Filos, Eszter Vértes, Zita Marinho, Gregory Farquhar, Diana Borsa, Abram Friesen, Feryal Behbahani, Tom Schaul, Andre Barreto, Simon Osindero

ICML 2022 · Reinforcement learning › Policy gradient and actor-critic · approach: Temporal-difference and discounting theory · 0 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 the IVE disagreement bonus to Dreamer planningDreamer · Walker Walk, η=0.3episodic return: 570±60 -> 1000±00Improved
Add the IVE disagreement bonus to Dreamer planningDreamer · Walker Walk, η=0.5episodic return: 80±50 -> 330±70Improved
Use the IVE mean rather than the base policy for planning (ablation)Dreamer · Walker Walk, η=0.2episodic return: 720±10 -> 860±40Improved
Replace the IVE mean with the 1-MPV (ablation)VPN(5) · Asterixepisodic return: 18.3±0.2 -> 15.1±0.6Hurt
Replace the IVE mean with the 5-MPV (ablation)VPN(5) · Asterixepisodic return: 18.3±0.2 -> 7.1±2.3Hurt
Replace the IVE mean with the 1-MPV (ablation)VPN(5) · Breakoutepisodic return: 22.0±0.7 -> 13.8±0.8Hurt
Replace the IVE mean with the 5-MPV (ablation)VPN(5) · Breakoutepisodic return: 22.0±0.7 -> 4.2±2.3Hurt
Replace the IVE mean with the 1-MPV (ablation)VPN(5) · Freewayepisodic return: 49.4±0.5 -> 49.1±0.7No clear effect
Replace the IVE mean with the 5-MPV (ablation)VPN(5) · Freewayepisodic return: 49.4±0.5 -> 24.3±4.2Hurt
Replace the IVE mean with the 1-MPV (ablation)VPN(5) · Seaquestepisodic return: 8.6±0.3 -> 4.7±0.9Hurt
Replace the IVE mean with the 5-MPV (ablation)VPN(5) · Seaquestepisodic return: 8.6±0.3 -> 1.2±1.4Hurt
Use self-inconsistency to seek novel statesTabular gridworld policy · 5×5 gridworldprobability of reaching the novel stateImproved

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