mlsanityData to 14 Oct 2026

Self-Improving In-Context Learning

arXiv 2605.23180 ↗First posted 22 May 2026cs.CL cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

We propose to improve in-context learning (ICL) by optimizing the continuous embeddings of a fixed few-shot prompt at test time. The key observation is that the log-probabilities a model assigns to its demonstrated outputsunicodex2013available from a single forward pass without generating any tokensunicodex2013provide a meaningful signal for how well the model has inferred the task from its demonstrations. We formalize this signal as a bounded, self-supervised confidence proxy and maximize it via zeroth-order optimization over the prompt embeddings, yielding a test-time calibration procedure. The approach requires no finetuning, no token generation, no predefined label set, and no external data, making it equally applicable to both classification and free-form generation tasks. Across a comprehensive suite of ICL tasks, the proposed calibration consistently matches or improves upon the base model and outperforms classification-specific baselines on most tasks. The statistically significant correlation between proxy improvement and downstream accuracy gain confirms that the proposed proxy encodes a reliable optimization signal for in-context learning.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 48 of its 58 references that are in the map. First posted on arXiv 22 May 2026.

Builds on

48 of 58 references.

References · 48

  1. 2017Attention Is All You Needcited by 60,361
  2. 2020Language Models are Few-Shot Learnerscited by 32,419
  3. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  4. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
  5. 2022Rethinking the Role of Demonstrations: What Makes In-Context Learning Work?cited by 1,434
  6. 2021Calibrate Before Use: Improving Few-Shot Performance of Language Modelscited by 1,392
  7. 2021What Makes Good In-Context Examples for GPT-3?cited by 1,195
  8. 2021Fantastically Ordered Prompts and Where to Find Them: Overcoming Few-Shot Prompt Order Sensitivitycited by 1,135
  9. 2022Transformers learn in-context by gradient descentcited by 732
  10. 2023Large Language Models Are Not Robust Multiple Choice Selectorscited by 486
  11. 2022Complexity-Based Prompting for Multi-Step Reasoningcited by 388
  12. 2023Fine-Tuning Language Models with Just Forward Passescited by 345
36 more
  1. 2023In-Context Learning Creates Task Vectorscited by 312
  2. 2023Function Vectors in Large Language Modelscited by 308
  3. 2021Surface Form Competition: Why the Highest Probability Answer Isn't Always Rightcited by 263
  4. 2023In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steeringcited by 245
  5. 2022Selective Annotation Makes Language Models Better Few-Shot Learnerscited by 243
  6. 2023Label Words are Anchors: An Information Flow Perspective for Understanding In-Context Learningcited by 236
  7. 2021Noisy Channel Language Model Prompting for Few-Shot Text Classificationcited by 165
  8. 2022Self-Adaptive In-Context Learning: An Information Compression Perspective for In-Context Example Selection and Orderingcited by 161
  9. 2022Diverse Demonstrations Improve In-context Compositional Generalizationcited by 139
  10. 2022Ground-Truth Labels Matter: A Deeper Look into Input-Label Demonstrationscited by 116
  11. 2023Finding Support Examples for In-Context Learningcited by 104
  12. 2023Batch Calibration: Rethinking Calibration for In-Context Learning and Prompt Engineeringcited by 94
  13. 2023Mitigating Label Biases for In-context Learningcited by 89
  14. 2023In-Context Learning with Iterative Demonstration Selectioncited by 74
  15. 2022Z-ICL: Zero-Shot In-Context Learning with Pseudo-Demonstrationscited by 70
  16. 2022Self-Generated In-Context Learning: Leveraging Auto-regressive Language Models as a Demonstration Generatorcited by 68
  17. 2022Prototypical Calibration for Few-shot Learning of Language Modelscited by 58
  18. 2024Revisiting Demonstration Selection Strategies in In-Context Learningcited by 58
  19. 2024Sparse MeZO: Less Parameters for Better Performance in Zeroth-Order LLM Fine-Tuningcited by 54
  20. 2024Variance-reduced Zeroth-Order Methods for Fine-Tuning Language Modelscited by 54
  21. 2023Better Zero-Shot Reasoning with Self-Adaptive Promptingcited by 52
  22. 2023kNN Prompting: Beyond-Context Learning with Calibration-Free Nearest Neighbor Inferencecited by 47
  23. 2024COME: Test-time adaption by Conservatively Minimizing Entropycited by 31
  24. 2023DPZero: Private Fine-Tuning of Language Models without Backpropagationcited by 26
  25. 2024Batch-ICL: Effective, Efficient, and Order-Agnostic In-Context Learningcited by 20
  26. 2023Universal Self-Adaptive Promptingcited by 14
  27. 2024Misconfidence-based Demonstration Selection for LLM In-Context Learningcited by 14
  28. 2024In-Context Example Ordering Guided by Label Distributionscited by 12
  29. 2024ICLEval: Evaluating In-Context Learning Ability of Large Language Modelscited by 12
  30. 2025Learning Task Representations from In-Context Learningcited by 11
  31. 2024Token-based Decision Criteria Are Suboptimal in In-context Learningcited by 9
  32. 2024Se2: Sequential Example Selection for In-Context Learningcited by 8
  33. 2025OptiSeq: Ordering Examples On-The-Fly for In-Context Learningcited by 6
  34. 2026Test-Time Detoxification without Training or Learning Anythingcited by 2
  35. 2026Test-Time Safety Alignmentcited by 2
  36. 2025D.Va: Validate Your Demonstration First Before You Use Itcited by 1

Built on by

Not cited yet.