AI Paper Atlas

Research map / Probabilistic and scientific modeling / Variational and generative sampling

MAVRL: Learning Reward Functions from Multiple Feedback Types with Amortized Variational Inference

Raphaël Baur, Yannick Metz, Maria Gkoulta, Mennatallah El-Assady, Giorgia Ramponi, Thomas Kleine Buening

ICML 2026 · Probabilistic and scientific modeling › Variational and generative sampling · approach: Amortized and online variational inference · 0 citations (OpenAlex)

Read the paper

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 stop feedback to demonstrations (ablation)MAVRL · grid_sparsenormalized return: 95.0 -> 100.0Improved
combine feedback types jointly instead of post-hoc averagingMAVRL · grid_cliffnormalized return: 54.9 -> 100.0Improved
combine feedback types jointly instead of post-hoc averagingMAVRL · LunarLander-v3normalized return: 59.9 -> 93.5Improved
use amortized variational inference instead of MCMCMAVRL · grid environmentswall time: 156.0 s -> 4.9 s on grid_cliffImproved
use amortized variational inference instead of MCMCMAVRL · grid_trapnormalized return: 93.0 -> 95.3No clear effect
combine all four feedback types instead of single modalityMAVRL · LunarLander-v3normalized return: 115.7 -> 93.5Hurt
combine all four feedback types instead of single modalityMAVRL · grid_cliffEPIC distance: lowest EPIC 0.481Improved
add multi-type feedback under preference misspecificationMAVRL · grid_cliffretained performance ratio: 0.19x -> 0.90xImproved
add multi-type feedback when demonstrations are corruptedMAVRL · grid_cliff and grid_sparseretained performance ratio: 0.38-0.50 ratiosHurt
use equal total feedback budget across modalities (ablation)MAVRL · tabular gridsnormalized return: single informative modality at full budget often beats multi-modal allocationMixed
use equal total feedback budget across modalities (ablation)MAVRL · continuous-controlnormalized returnImproved
combine feedback types instead of imitation under dynamics perturbationMAVRL · LunarLander-v3 gravity and windnormalized returnImproved

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