few-shot

Meta-iLaD 提出一种通过跨动力学环境元学习实现可识别潜在动力学的新框架，利用少量样本的条件先验快速提取环境变量，并同时对潜在状态、环境变量与动力学函数建立可识别性，从而提升对未见环境的泛化与预测能力。

提出一种通过蒸馏最优传输在仅有100-500个配对样本下对齐两个独立预训练单模态编码器的方法，在Flickr30k、MS-COCO和ChEBI-20上以相同锚点预算超越现有方法。

Goldsmith, an agentic annotation pipeline presented in an arXiv paper, converts small expert-annotated gold sets into structured annotation definitions optimized via an executable loss-guided loop, outperforming direct rewriting, OPRO, APE, and PromptBreeder.

This paper evaluates retrieval-based in-context learning approaches for detecting criminally relevant hate speech in German social media posts, finding that few-shot prompting outperforms zero-shot but retrieval methods offer marginal gains, with models better suited for triage than autonomous moderation.

The paper introduces TPGC, a dual-prior prompt initialization method for multi-task graph pre-training that combines task and structural priors to improve alignment and transferability, achieving superior performance in few-shot scenarios.

SkalskiP highlights Qwen3.8-Max, a vision-language model for object detection that can be prompted with positive and negative boxes to generate detections, achieving 60-80% mAP with single or multiple prompts and performing well on diverse image types.

Proposes DuPLeR, a dual-path LLM reasoning framework for multimodal few-shot knowledge graph completion, combining LLM-derived type priors with factual structures to improve inductive KGC under data scarcity.

This paper introduces cross-domain off-policy evaluation and learning (OPE/L) for contextual bandits, allowing the use of logged data from multiple source domains to improve policy evaluation and learning in target domains with challenging conditions like few-shot data, deterministic logging policies, and new actions.

Introduces DriveDNA, a large-scale multimodal naturalistic driving dataset with 4,121 drives from 465 drivers across 115 vehicle models, and a benchmark for driving style identification through tasks like few-shot driver re-identification and personalized behavior prediction.

This paper studies how LLM adaptation strategies (definitions, examples, fine-tuning) transfer under source shift in climate disclosure classification, finding that simpler strategies like definitions transfer more consistently than complex ones.

This paper presents a comprehensive evaluation of five large language models for citation function classification, achieving new state-of-the-art results on the ACL-ARC dataset with a fine-tuned Falcon 7B model. It also introduces the AC3 dataset, which includes a seven-category annotation scheme distinguishing neutral acknowledgments from evaluative stances.

Fable 5 model only used 4 prompts and $173 worth of tokens to create a game called 'Super Smart Racing', demonstrating its extremely strong generative capabilities.

This paper challenges the assumption that reranking always improves few-shot selection for LLMs, proposing a training-free gated reranking approach that uses model uncertainty to decide when to rerank, reducing computational costs by 15-80% while slightly improving performance.

This paper compares fine-tuned BERT (gbert-large) with few-shot LLM prompting (Llama 4 Maverick) for detecting threat and solution framing in German climate news sentences. BERT achieves higher F1 scores (0.83 vs 0.78), and an ablation study shows that providing preceding sentence context improves performance.

Introduces AnySimLite, a lightweight similarity encoder for on-device speech-adjacent classification tasks, achieving state-of-the-art or competitive performance while using less than 1/250th the model size of the qLLaMA-LoRA-7B baseline.

This study investigates whether instruction-tuned LLMs (Llama-3.1-8B, Qwen2.5-7B, Mistral-7B, Phi-3-mini) can reliably classify Correct Information Units in aphasic discourse transcripts. Few-shot prompting yields competitive F1 scores (0.776–0.817) for three models, but performance varies by severity and human agreement remains insufficient for fully autonomous use.

This paper investigates few-shot biomedical relation extraction using prompt-based learning with LLMs, comparing pairwise classification and joint generation approaches. The best model achieves micro-F1 of 0.44, outperforming previous few-shot results but remaining below supervised baselines, while macro-F1 surpasses the supervised baseline on rare relation types.

PrintGuard 2.0 is a major rewrite of a few-shot FDM fault detector using a ShuffleNetV2 backbone and prototypical network, now with a single Python engine that runs unmodified on both CPython and Pyodide in the browser via a platform abstraction layer, enabling per-printer sensitivity tuning and fair inference scheduling.

This paper proposes LLM-GNN Co-Teaching, a bidirectional framework for few-shot graph learning on text-attributed graphs. The LLM and GNN exchange confident pseudo-labels and use round-based preference optimization (RPL-PO) to mutually improve, outperforming prior methods on benchmarks.

Proposes Demo2Reward, a test-time prompt optimization technique for VLM reward models using a few expert demonstrations, significantly reducing false positives and improving policy learning in robotics without additional model training.