mlsanityData to 7 Oct 2026

Combining Induction and Transduction for Abstract Reasoning

arXiv 2411.02272 ↗First posted 4 Nov 2024cs.LG cs.AI cs.CLcited by 53

On the map: LLM reasoning → Multimodal reasoning

Abstract (arXiv)

When learning an input-output mapping from very few examples, is it better to first infer a latent function that explains the examples, or is it better to directly predict new test outputs, e.g. using a neural network? We study this question on ARC by training neural models for induction (inferring latent functions) and transduction (directly predicting the test output for a given test input). We train on synthetically generated variations of Python programs that solve ARC training tasks. We find inductive and transductive models solve different kinds of test problems, despite having the same training problems and sharing the same neural architecture: Inductive program synthesis excels at precise computations, and at composing multiple concepts, while transduction succeeds on fuzzier perceptual concepts. Ensembling them approaches human-level performance on ARC.

Key facts. Cited by 53 papers in the map (data through 7 Oct 2026). Placed in Multimodal reasoning, one of 17 directions in LLM reasoning. Builds on 31 of its 62 references that are in the map. First posted on arXiv 4 Nov 2024.

Builds on

31 of 62 references.

Closest ancestors · 3

Same direction, newest first

  1. 2024Do Large Language Models Solve ARC Visual Analogies Like People Do?13 Mar
  2. 2024CodeIt: Self-Improving Language Models with Prioritized Hindsight Replay7 Feb
  3. 2023Solving ARC visual analogies with neural embeddings and vector arithmetic: A generalized method14 Nov

Other references · 28

  1. 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,160
  2. 2021Evaluating Large Language Models Trained on Codecited by 8,053
  3. 2021Program Synthesis with Large Language Modelscited by 3,509
  4. 2022Self-Instruct: Aligning Language Models with Self-Generated Instructionscited by 2,674
  5. 2022Competition-Level Code Generation with AlphaCodecited by 1,427
  6. 2022Code as Policies: Language Model Programs for Embodied Controlcited by 1,172
  7. 2022PAL: Program-aided Language Modelscited by 635
  8. 2017RobustFill: Neural Program Learning under Noisy I/Ocited by 124
  9. 2023Chain of Code: Reasoning with a Language Model-Augmented Code Emulatorcited by 120
  10. 2023Hypothesis Search: Inductive Reasoning with Language Modelscited by 119
  11. 2024WorldCoder, a Model-Based LLM Agent: Building World Models by Writing Code and Interacting with the Environmentcited by 99
  12. 2023The ConceptARC Benchmark: Evaluating Understanding and Generalization in the ARC Domaincited by 97
16 more
  1. 2015Neural Programmer-Interpreterscited by 96
  2. 2022Top-Down Synthesis for Library Learningcited by 57
  3. 2018Modular meta-learningcited by 51
  4. 2024Code Repair with LLMs gives an Exploration-Exploitation Tradeoffcited by 50
  5. 2024VisualPredicator: Learning Abstract World Models with Neuro-Symbolic Predicates for Robot Planningcited by 37
  6. 2023Learning adaptive planning representations with natural language guidancecited by 36
  7. 2024Reasoning Abilities of Large Language Models: In-Depth Analysis on the Abstraction and Reasoning Corpuscited by 32
  8. 2024H-ARC: A Robust Estimate of Human Performance on the Abstraction and Reasoning Corpus Benchmarkcited by 32
  9. 2016Inference Compilation and Universal Probabilistic Programmingcited by 31
  10. 2021Neural-guided, Bidirectional Program Search for Abstraction and Reasoningcited by 23
  11. 2016Deep Amortized Inference for Probabilistic Programscited by 22
  12. 2022From Perception to Programs: Regularize, Overparameterize, and Amortizecited by 13
  13. 2023ANPL: Towards Natural Programming with Interactive Decompositioncited by 13
  14. 2023A Neurodiversity-Inspired Solver for the Abstraction & Reasoning Corpus (ARC) Using Visual Imagery and Program Synthesiscited by 9
  15. 2023A Divide-Align-Conquer Strategy for Program Synthesiscited by 7
  16. 2024Relational decomposition for program synthesiscited by 6

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 3; 2025: 30; 2026: 20.

53 papers.

Sort:

By area:

  1. 2025Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Modelscited by 381
  2. 2024ARC Prize 2024: Technical Reportcited by 103
  3. 2024The Surprising Effectiveness of Test-Time Training for Few-Shot Learningcited by 88
  4. 2025Revisiting Reinforcement Learning for LLM Reasoning from A Cross-Domain Perspectivecited by 75
  5. 2025ArcMemo: Abstract Reasoning Composition with Lifelong LLM Memorycited by 22
  6. 2025RLAD: Training LLMs to Discover Abstractions for Solving Reasoning Problemscited by 21
  7. 2025Is That Your Final Answer? Test-Time Scaling Improves Selective Question Answeringcited by 17
  8. 2025Product of Experts with LLMs: Boosting Performance on ARC Is a Matter of Perspectivecited by 17
  9. 2026LoopViT: Scaling Visual ARC with Looped Transformerscited by 14
  10. 2025Generating Computational Cognitive Models using Large Language Modelscited by 12
  11. 2025BOTS: A Unified Framework for Bayesian Online Task Selection in LLM Reinforcement Finetuningcited by 12
  12. 2025EasyARC: Evaluating Vision Language Models on True Visual Reasoningcited by 10
38 more
  1. 2025ARC-GEN: A Mimetic Procedural Benchmark Generator for the Abstraction and Reasoning Corpuscited by 8
  2. 2025MIR-Bench: Can Your LLM Recognize Complicated Patterns via Many-Shot In-Context Reasoning?cited by 7
  3. 2025Shoot First, Ask Questions Later? Building Rational Agents that Explore and Act Like Peoplecited by 7
  4. 2025Diverse Inference and Verification for Advanced Reasoningcited by 6
  5. 2025MiGrATe: Mixed-Policy GRPO for Adaptation at Test-Timecited by 6
  6. 2025Think Visually, Reason Textually: Vision-Language Synergy in ARCcited by 6
  7. 2026Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world modelscited by 6
  8. 2025Your Reasoning Benchmark May Not Test Reasoning: Revealing Perception Bottleneck in Abstract Reasoning Benchmarkscited by 5
  9. 2025Benchmarking Abstract and Reasoning Abilities Through A Theoretical Perspectivecited by 4
  10. 2025CausalARC: Abstract Reasoning with Causal World Modelscited by 4
  11. 2025Program Synthesis via Test-Time Transductioncited by 4
  12. 2026Transferability for General Reasoning: An Automated Curriculum for Multi-Domain RLVRcited by 4
  13. 2026Twin: Playing an Unknown Game with a Test-Time Digital Twincited by 4
  14. 2025ARC-AGI Without Pretrainingcited by 3
  15. 2026Symbol-Equivariant Recurrent Reasoning Modelscited by 3
  16. 2025Shedding Light in Task Decomposition in Program Synthesis: The Driving Force of the Synthesizer Modelcited by 2
  17. 2025Cognitive maps are generative programscited by 2
  18. 2025From Reasoning to Generalization: Knowledge-Augmented LLMs for ARC Benchmarkcited by 2
  19. 2025Beyond Either-Or Reasoning: Transduction and Induction as Cooperative Problem-Solving Paradigmscited by 1
  20. 2025GIFARC: Synthetic Dataset for Leveraging Human-Intuitive Analogies to Elevate AI Reasoningcited by 1
  21. 2025Bridging Reasoning to Learning: Unmasking Illusions using Complexity Out of Distribution Generalizationcited by 1
  22. 2025Vector Symbolic Algebras for the Abstraction and Reasoning Corpuscited by 1
  23. 2026Bongards at the Boundary of Perception and Reasoning: Programs or Language?cited by 1
  24. 2026Playing ZendoWorld: Challenging AI Agents on Active Visual Concept Inductioncited by 1
  25. 2026TraceViT: Grounded Trace Supervision for Visual Abstract Reasoningcited by 1
  26. 2026Beyond Solvability: Task Learnability as a Static Prior for LLM RL Post-Trainingcited by 1
  27. 2026Implicit Rule Induction with Test-Time Task Embeddings in ARC-like Taskscited by 1
  28. 2024Pattern Analogies: Learning to Perform Programmatic Image Edits by Analogycited by 0
  29. 2025Channel-Wise MLPs Improve the Generalization of Recurrent Convolutional Networkscited by 0
  30. 2025Resource Allocation in Hybrid Radio-Optical IoT Networks using GNN with Multi-task Learningcited by 0
  31. 2025From Priors to Predictions: Explaining and Visualizing Human Reasoning in a Graph Neural Network Frameworkcited by 0
  32. 2026Transformer See, Transformer Do: Copying as an Intermediate Step in Learning Analogical Reasoningcited by 0
  33. 2026Modality-Driven Search with Holistic Trace Judging for ARC-AGI-2cited by 0
  34. 2026BDH-CQ: In-Context Learning with Recurrent Latent Reasoningcited by 0
  35. 2026Position: Reasoning is a Learnable Rule-Based Processcited by 0
  36. 2026Narcissus: Program Synthesis Using Context-Aware LLM Approximationscited by 0
  37. 2026World-Time Compute with Verified Code World Modelscited by 0
  38. 2026Exactness at Inference: A Representational Criterion for Out-of-Distribution Generalizationcited by 0

50 most cited of 53 All 53 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024Neural networks for abstraction and reasoning: Towards broad generalization in machines
  2. 2024Untrained neural networks can demonstrate memorization-independent abstract reasoning
  3. 2024Program Synthesis using Inductive Logic Programming for the Abstraction and Reasoning Corpus
  4. 2025NSA: Neuro-symbolic ARC Challenge
  5. 2024Grokked Transformers are Implicit Reasoners: A Mechanistic Journey to the Edge of Generalization
  6. 2026In-Context Learning Amplifies a Latent Symbolic Circuit
  7. 2025Implicit Reasoning in Transformers is Reasoning through Shortcuts
  8. 2026A Systematic Study of Small Language Models on Abstract Reasoning Tasks