diagnostic metrics 1latent dynamics 1model-based reinforcement learning 1value equivalence 1world model evaluation 1
From the 1 of 3 linked papers with an AI index.
3 papers
cs.LG2026
Operator-on-F complements value-equivalence: a planning-time diagnostic for latent world models
Donna Vakalis
The paper introduces a new diagnostic called operator-on-F that measures how well a latent world model’s predictions match the environment on a chosen observable set, showing stron…
cs.LG2026
What a World Model Represents Is Three Questions
Donna Vakalis
World models learn task-relevant information through many routes: observation reconstruction, recurrent state, temporal filtering, and explicit task supervision. Different routes c…
cs.LG2026
In-Context Reinforcement Learning through Bayesian Fusion of Context and Value Prior
Anaïs Berkes, Vincent Taboga, Donna Vakalis +2
In-context reinforcement learning (ICRL) promises fast adaptation to unseen environments without parameter updates, but current methods either cannot improve beyond the training di…