mlsanityData to 7 Oct 2026

Meta-trained agents implement Bayes-optimal agents

arXiv 2010.11223 ↗First posted 21 Oct 2020cs.AI cs.LG cs.NEcited by 37

Not on the map. Why?

Abstract (arXiv)

Memory-based meta-learning is a powerful technique to build agents that adapt fast to any task within a target distribution. A previous theoretical study has argued that this remarkable performance is because the meta-training protocol incentivises agents to behave Bayes-optimally. We empirically investigate this claim on a number of prediction and bandit tasks. Inspired by ideas from theoretical computer science, we show that meta-learned and Bayes-optimal agents not only behave alike, but they even share a similar computational structure, in the sense that one agent system can approximately simulate the other. Furthermore, we show that Bayes-optimal agents are fixed points of the meta-learning dynamics. Our results suggest that memory-based meta-learning might serve as a general technique for numerically approximating Bayes-optimal agents - that is, even for task distributions for which we currently don't possess tractable models.

Key facts. Cited by 37 papers in the map (data through 7 Oct 2026). First posted on arXiv 21 Oct 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 3; 2022: 8; 2023: 7; 2024: 8; 2025: 4; 2026: 7.

37 papers.

Sort:

By area:

  1. 2023Human-Timescale Adaptation in an Open-Ended Task Spacecited by 136
  2. 2022General-Purpose In-Context Learning by Meta-Learning Transformerscited by 102
  3. 2021Alchemy: A benchmark and analysis toolkit for meta-reinforcement learning agentscited by 35
  4. 2024LMAct: A Benchmark for In-Context Imitation Learning with Long Multimodal Demonstrationscited by 32
  5. 2024Learning Universal Predictorscited by 29
  6. 2023Bayes in the age of intelligent machinescited by 22
  7. 2023Memory-Based Meta-Learning on Non-Stationary Distributionscited by 14
  8. 2024In-context learning agents are asymmetric belief updaterscited by 13
  9. 2022Recurrent networks, hidden states and beliefs in partially observable environmentscited by 11
  10. 2024What Should Embeddings Embed? Autoregressive Models Represent Latent Generating Distributionscited by 10
  11. 2022Learning Robust Real-Time Cultural Transmission without Human Datacited by 9
  12. 2022Simplex Neural Population Learning: Any-Mixture Bayes-Optimality in Symmetric Zero-sum Gamescited by 8
25 more
  1. 2021Model-Free Risk-Sensitive Reinforcement Learningcited by 7
  2. 2022Learning Causal Overhypotheses through Exploration in Children and Computational Modelscited by 7
  3. 2024In-context learning and Occam's razorcited by 7
  4. 2025Understanding Prompt Tuning and In-Context Learning via Meta-Learningcited by 6
  5. 2023Minimax-Bayes Reinforcement Learningcited by 5
  6. 2023MONGOOSE: Path-wise Smooth Bayesian Optimisation via Meta-learningcited by 5
  7. 2024Transformers for Supervised Online Continual Learningcited by 5
  8. 2022Planning to the Information Horizon of BAMDPs via Epistemic State Abstractioncited by 4
  9. 2024BAMDP Shaping: a Unified Framework for Intrinsic Motivation and Reward Shapingcited by 4
  10. 2025Why is prompting hard? Understanding prompts on binary sequence predictorscited by 4
  11. 2024On Uncertainty Quantification for Near-Bayes Optimal Algorithmscited by 3
  12. 2026Position: agentic AI orchestration should be Bayes-consistentcited by 3
  13. 2026ATLAS: Active Theory Learning for Automated Sciencecited by 3
  14. 2021Emergent behavior and neural dynamics in artificial agents tracking turbulent plumescited by 2
  15. 2023Doing the right thing for the right reason: Evaluating artificial moral cognition by probing cost insensitivitycited by 2
  16. 2022Modeling Human Behavior Part I -- Learning and Belief Approachescited by 1
  17. 2022Beyond Bayes-optimality: meta-learning what you know you don't knowcited by 1
  18. 2023A Metalearned Neural Circuit for Nonparametric Bayesian Inferencecited by 1
  19. 2025Bayesian Meta-Reinforcement Learning with Laplace Variational Recurrent Networkscited by 1
  20. 2026What does a Bayes-filtered transformer believe? A predictive Monte Carlo approachcited by 1
  21. 2025Predictive Coding Enhances Meta-RL To Achieve Interpretable Bayes-Optimal Belief Representation Under Partial Observabilitycited by 0
  22. 2026VariBASed: Variational Bayes-Adaptive Sequential Monte-Carlo Planning for Deep Reinforcement Learningcited by 0
  23. 2026Meta-Learning and Meta-Reinforcement Learning -- Tracing the Path towards DeepMind's Adaptive Agentcited by 0
  24. 2026Can In-Context Learning Support Intrinsic Curiosity?cited by 0
  25. 2026Elicitation without Backpropagation: Steering Model Behavior by Optimizing the Latent Posteriorcited by 0