mlsanityData to 7 Oct 2026

Using Natural Language and Program Abstractions to Instill Human Inductive Biases in Machines

arXiv 2205.11558 ↗First posted 23 May 2022cs.AIcited by 20

Not on the map. Why?

Abstract (arXiv)

Strong inductive biases give humans the ability to quickly learn to perform a variety of tasks. Although meta-learning is a method to endow neural networks with useful inductive biases, agents trained by meta-learning may sometimes acquire very different strategies from humans. We show that co-training these agents on predicting representations from natural language task descriptions and programs induced to generate such tasks guides them toward more human-like inductive biases. Human-generated language descriptions and program induction models that add new learned primitives both contain abstract concepts that can compress description length. Co-training on these representations result in more human-like behavior in downstream meta-reinforcement learning agents than less abstract controls (synthetic language descriptions, program induction without learned primitives), suggesting that the abstraction supported by these representations is key.

Key facts. Cited by 20 papers in the map (data through 7 Oct 2026). Builds on 34 of its 63 references that are in the map. First posted on arXiv 23 May 2022.

Builds on

34 of 63 references.

References · 34

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  3. 2017Proximal Policy Optimization Algorithmscited by 14,441
  4. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,266
  5. 2019Sentence-BERT: Sentence Embeddings using Siamese BERT-Networkscited by 9,049
  6. 2018Relational inductive biases, deep learning, and graph networkscited by 1,341
  7. 2020Meta-Learning in Neural Networks: A Surveycited by 944
  8. 2021Multimodal Few-Shot Learning with Frozen Language Modelscited by 657
  9. 2016Reinforcement Learning with Unsupervised Auxiliary Taskscited by 400
  10. 2016Learning to reinforcement learncited by 369
  11. 2019The Neuro-Symbolic Concept Learner: Interpreting Scenes, Words, and Sentences From Natural Supervisioncited by 369
  12. 2020Inductive Biases for Deep Learning of Higher-Level Cognitioncited by 273
22 more
  1. 2018Machine Theory of Mindcited by 222
  2. 2016Learning to Navigate in Complex Environmentscited by 192
  3. 2019A Survey of Reinforcement Learning Informed by Natural Languagecited by 154
  4. 2017Grounded Language Learning in a Simulated 3D Worldcited by 100
  5. 2017Grounding Referring Expressions in Images by Variational Contextcited by 82
  6. 2017Learning with Latent Languagecited by 75
  7. 2020Learning Compositional Rules via Neural Program Synthesiscited by 74
  8. 2017Learning to Infer Graphics Programs from Hand-Drawn Imagescited by 73
  9. 2016Loss is its own Reward: Self-Supervision for Reinforcement Learningcited by 72
  10. 2019Meta-learning of Sequential Strategiescited by 68
  11. 2019Causal Reasoning from Meta-reinforcement Learningcited by 53
  12. 2019Environmental drivers of systematicity and generalization in a situated agentcited by 50
  13. 2021Leveraging Language to Learn Program Abstractions and Search Heuristicscited by 44
  14. 2021Tell me why! Explanations support learning relational and causal structurecited by 36
  15. 2020Universal linguistic inductive biases via meta-learningcited by 28
  16. 2020Gibbs Sampling with Peoplecited by 28
  17. 2019Shaping Visual Representations with Language for Few-shot Classificationcited by 27
  18. 2020Learning abstract structure for drawing by efficient motor program inductioncited by 23
  19. 2021Learning to communicate about shared procedural abstractionscited by 15
  20. 2022Identifying concept libraries from language about object structurecited by 8
  21. 2022Disentangling Abstraction from Statistical Pattern Matching in Human and Machine Learningcited by 6
  22. 2018Guided Feature Transformation (GFT): A Neural Language Grounding Module for Embodied Agentscited by 5

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 1; 2023: 10; 2024: 6; 2025: 2; 2026: 1.

20 papers.

Sort:

By area:

  1. 2023Getting aligned on representational alignmentcited by 142
  2. 2024Scaling Instructable Agents Across Many Simulated Worldscited by 67
  3. 2023Language Instructed Reinforcement Learning for Human-AI Coordinationcited by 61
  4. 2023The SocialAI School: Insights from Developmental Psychology Towards Artificial Socio-Cultural Agentscited by 24
  5. 2024Doing Experiments and Revising Rules with Natural Language and Probabilistic Reasoningcited by 23
  6. 2023Human-like Few-Shot Learning via Bayesian Reasoning over Natural Languagecited by 21
  7. 2023Structure in Deep Reinforcement Learning: A Survey and Open Problemscited by 16
  8. 2023Reinforcement Learning with Simple Sequence Priorscited by 11
  9. 2024Seeing Like an AI: How LLMs Apply (and Misapply) Wikipedia Neutrality Normscited by 8
  10. 2023Predictive auxiliary objectives in deep RL mimic learning in the braincited by 7
  11. 2023Instilling Inductive Biases with Subnetworkscited by 6
  12. 2023Hierarchical reinforcement learning with natural language subgoalscited by 3
8 more
  1. 2024Human-like Category Learning by Injecting Ecological Priors from Large Language Models into Neural Networkscited by 3
  2. 2024Comparing Abstraction in Humans and Large Language Models Using Multimodal Serial Reproductioncited by 2
  3. 2025Cognitive maps are generative programscited by 2
  4. 2022Leveraging Systematic Knowledge of 2D Transformationscited by 1
  5. 2023Do as I can, not as I getcited by 1
  6. 2025Metareasoning constraints couple narratives, affect and cognitioncited by 1
  7. 2024FairBelief -- Assessing Harmful Beliefs in Language Modelscited by 0
  8. 2026Prospective Compression in Human Abstraction Learningcited by 0