Looped linear transformers with LN provably converge via GD to implement the power method on principal component prediction.
hub Canonical reference
A Survey on In-context Learning
Canonical reference. 93% of citing Pith papers cite this work as background.
78
Pith papers citing it
Background
93% of classified citations
abstract
With the increasing capabilities of large language models (LLMs), in-context learning (ICL) has emerged as a new paradigm for natural language processing (NLP), where LLMs make predictions based on contexts augmented with a few examples. It has been a significant trend to explore ICL to evaluate and extrapolate the ability of LLMs. In this paper, we aim to survey and summarize the progress and challenges of ICL. We first present a formal definition of ICL and clarify its correlation to related studies. Then, we organize and discuss advanced techniques, including training strategies, prompt designing strategies, and related analysis. Additionally, we explore various ICL application scenarios, such as data engineering and knowledge updating. Finally, we address the challenges of ICL and suggest potential directions for further research. We hope that our work can encourage more research on uncovering how ICL works and improving ICL.
hub tools
citation-role summary
background 29
method 1
citation-polarity summary
claims ledger
- abstract With the increasing capabilities of large language models (LLMs), in-context learning (ICL) has emerged as a new paradigm for natural language processing (NLP), where LLMs make predictions based on contexts augmented with a few examples. It has been a significant trend to explore ICL to evaluate and extrapolate the ability of LLMs. In this paper, we aim to survey and summarize the progress and challenges of ICL. We first present a formal definition of ICL and clarify its correlation to related studies. Then, we organize and discuss advanced techniques, including training strategies, prompt des
co-cited works
representative citing papers
NLI autonomously discovers a vocabulary of primitive operations and interprets variable-length programs via a neural executor, allowing end-to-end training and gradient-based test-time adaptation that outperforms prior methods on combinatorial generalization tasks.
VASO is a verification-guided self-evolution framework for LLM robot skill contracts that reaches 97.2% formal-specification compliance on Jackal and quadcopter tasks using under 100 samples.
WALDO reformulates zero-shot anomaly localisation as comparative inference using entropy-weighted Sliced Wasserstein distances for reference selection and Goldilocks zone sampling, achieving 19% relative mAP@30 improvement on the NOVA brain MRI benchmark.
Pro²Assist uses multimodal egocentric perception from AR glasses to track fine-grained progress in long-horizon procedural tasks and deliver timely proactive assistance, outperforming baselines by over 21% in action understanding and up to 2.29x in timing accuracy.
AnchorSeg uses ordered query banks of latent reasoning tokens plus a spatial anchor token and a Token-Mask Cycle Consistency loss to achieve 67.7% gIoU and 68.1% cIoU on the ReasonSeg benchmark.
DeformPIC deforms query point clouds under prompt guidance for in-context learning, outperforming prior methods with lower Chamfer Distance on reconstruction, denoising, and registration tasks.
A new dataset and nine-metric majority-vote procedure show that existing code-reasoning benchmarks are dominated by lower-complexity problems that do not reflect real-world code.
ProAgent uses on-demand tiered perception and context-aware LLM reasoning to deliver proactive assistance on AR glasses, achieving up to 27.7% higher prediction accuracy and 20.5% lower false detections than baselines.
CREST-Search is a red-teaming framework that crafts seemingly benign search queries to induce unsafe citations from web-augmented LLMs, backed by a new WebSearch-Harm dataset for fine-tuning a specialized attacker model.
SITE applies soft gradient-based head selection to inject ICL-derived task embeddings, outperforming prior embedding adaptation and few-shot ICL across generation, reasoning, and NLU tasks on 12 LLMs from 4B to 70B parameters.
One training example via RLVR boosts LLM math reasoning from 17.6% to 35.7% average across six benchmarks.
CodeMind evaluates ten LLMs on four benchmarks using three new code reasoning tasks, finding performance varies by model size and drops with complexity while showing no correlation with bug repair ability.
Set-of-Mark prompting marks segmented image regions with alphanumerics and masks to let GPT-4V achieve state-of-the-art zero-shot results on referring expression comprehension and segmentation benchmarks like RefCOCOg.
Traj-Evolve combines non-parametric experience retrieval and multi-agent RL with a leave-one-out unification strategy to outperform baselines on lung cancer prediction from up to five years of multimodal EHRs, including in never-smokers.
CDS-trained BabyLMs show earlier and more appropriate production in a new frame-completion task while FineWeb-edu models lead on comprehension benchmarks, indicating current tests underestimate CDS benefits.
TCFDNet uses a Gait Modality Text Dictionary from LLMs, CLIP alignment, and text-guided disentanglement modules to achieve SOTA cross-modal gait recognition on SUSTech1K and FreeGait.
DIVE improves in-context vector distillation for medical report generation via decisive-token supervision on pathology terms and EOS plus state-conditioned dynamic steering, achieving top BLEU-4, ROUGE-L and RadGraph F1 on MIMIC-CXR and CheXpert Plus.
LLMs show instruction-following rates from 1% to 99% when instructions conflict with hardcoded pattern demonstrations, with output diversity as the main predictor of resistance.
METIS internalizes curriculum judgment in LLM reinforcement fine-tuning by predicting within-prompt reward variance via in-context learning and jointly optimizing with a self-judgment reward, yielding superior performance and up to 67% faster convergence across math, code, and agent benchmarks.
Introduces Personal VCL formalization and benchmark revealing LMM context gaps, plus an Agentic Context Bank baseline that boosts personalized visual reasoning.
Decompose and Recompose decomposes seen robotic demonstrations into skill-action alignments and recomposes them via visual-semantic retrieval and planning to enable zero-shot cross-task generalization.
Ctx2Skill uses a self-evolving multi-agent loop with Challenger, Reasoner, Judge, and Cross-time Replay to discover context-specific skills, improving task-solving rates on CL-bench benchmarks across models.
Retrieved query variants from logs combined with LLM-augmented generation improve unsupervised QPP accuracy by up to 30% for neural rankers on TREC DL'19 and DL'20.
citing papers explorer
-
Evaluating Code Reasoning Abilities of Large Language Models Under Real-World Settings
A new dataset and nine-metric majority-vote procedure show that existing code-reasoning benchmarks are dominated by lower-complexity problems that do not reflect real-world code.
-
BugScope: Learn to Find Bugs Like Human
BugScope structures LLM bug detection into three human-mirroring steps and distills guidelines from examples, reaching 0.87 F1 on 33 real bugs while outperforming Claude and Cursor tools and uncovering 184 new issues in production code.
-
Context-Guided Decompilation: A Step Towards Re-executability
ICL4Decomp applies in-context learning to guide LLMs in generating re-executable decompiled code from binaries, reporting roughly 40% higher re-executability than prior methods across datasets and optimization levels.