in-context-learning

Tag

Cards List
#in-context-learning

@garykbrixi: Language models can learn new tasks from just a few examples in their prompt, with no retraining. This is called in-con…

X AI KOLs Following ↗ · 2026-10-29 Cached

New work shows that genome language models like Evo 2 can perform in-context learning from a few examples, generating proteins, de novo antibodies, CRISPR-Cas9 loci, mitochondrial genomes, and even complete bacteriophages, with iterative chain-of-thought-style prompting for harder design goals.

0 favorites 0 likes
#in-context-learning

Collaborative In-Context Learning for Indoor Radio Map Estimation

arXiv cs.LG ↗ · 2026-10-27 Cached

This paper formulates indoor radio map estimation as a collaborative in-context learning problem and proposes Poly-Ripple, a framework where multiple clients use large autoregressive vision models without model updates, employing ripple-ordered autoregressive modeling and importance-aware aggregation to outperform ICL baselines.

0 favorites 0 likes
#in-context-learning

From Noisy Retrieval to Decision Evidence: Silver Evidence Demonstrations for In-Context Learning

arXiv cs.CL ↗ · 2026-10-26

提出 silver-evidence demonstrations 方法,用生成的、目标条件化的证据摘要替换 ICL 示例中的原始检索段落,以减少噪声检索带来的干扰。该方法在 EI 2.0、SciFact、合成策略检索和 HICRIC/IMR 数据集上优于 gold-only ICL 和 InstructRAG 式 ICL,并在所有标注数据集上提升了证据 F1。

0 favorites 0 likes
#in-context-learning

POS-Signature Demonstration Retrieval for In-Context Dependency Parsing

arXiv cs.CL ↗ · 2026-10-26

本论文提出一种 POS-signature 检索策略,用于依存句法分析的 in-context learning 示例选择,结果显示其在 UD English Web Treebank 测试集上的标注附着分数优于语义相似度检索基线。

0 favorites 0 likes
#in-context-learning

ReGE: Stronger Agents through Experience Synthesis

Hugging Face Daily Papers ↗ · 2026-10-24 Cached

ReGE 是一个将智能体历史执行经验合成为自指导信息的框架,使后续任务尝试无需更强教师模型或参数更新即可提升表现,在四个基准上平均提升约 9-10 个百分点,并大幅减少引导构建所需的 token 成本。

0 favorites 0 likes
#in-context-learning

@heyshrutimishra: SenseNova-RoboRSI shows why strong benchmark scores are not enough for embodied AI. Real robots face collisions, overhe…

X AI KOLs Following ↗ · 2026-10-22 Cached

A technical report evaluating GPT-6-Astra across 12 embodied AI benchmarks finds it achieves unprecedented manipulation and navigation performance, yet reveals large gaps—costly demonstrations, physical constraints, and irreversible failures—preventing reliable real-world deployment.

0 favorites 0 likes
#in-context-learning

ICLHerd: Herding-Based Data Selection for LLMs with In-Context Learning as a Proxy

arXiv cs.CL ↗ · 2026-10-21 Cached

ICCLHerd 提出一种基于 herding 的 LLM 训练数据选择框架,利用 von Mises–Fisher 核平衡 mode-seeking 与 mode-averaging,并以 in-context learning 作为验证代理进行超参数搜索,实验表明其优于现有数据选择算法。

0 favorites 0 likes
#in-context-learning

From Observation to Intervention: Distilling Self-Generated In-Context Demonstrations via Executable Causal Intervention

arXiv cs.CL ↗ · 2026-10-16 Cached

该论文提出 ECI2D 方法,通过将语言推理链抽象为可执行程序并施加因果干预(do-operator)来筛选自生成的上下文示例,解决仅靠观察过滤示例时的观测等价问题,在数学、逻辑与常识推理基准上取得显著提升。

0 favorites 0 likes
#in-context-learning

How to Evaluate When You Don't Know How to Evaluate? Choosing Reliable Setups for Base Language Model Benchmarking

arXiv cs.CL ↗ · 2026-10-15 Cached

This paper investigates how different evaluation setup choices affect base language model benchmarking by examining 8 design aspects across 26 2B-parameter models and larger LLMs in four languages, proposing monotonicity, non-randomness, and ranking consistency as meta-measures to guide reliable evaluation setup selection during pretraining.

0 favorites 0 likes
#in-context-learning

From Copying to Composition: How Language Models Develop In-Context Learning

arXiv cs.CL ↗ · 2026-10-15 Cached

本文提出一种受控诊断任务,区分语言模型在上下文学习中的“复制示例”与“组合应用规则”两种行为,发现提示控制、答案偏好、无候选构造和迁移能力在训练中沿不同时间线发展,且合成定义文档训练能提升组合式响应。

0 favorites 0 likes
#in-context-learning

CB-Agent: Concept-Based In-Context Learning for Sequential Action Selection

arXiv cs.AI ↗ · 2026-10-15 Cached

The paper proposes CB-Agent, a concept-based in-context learning method that fits a reusable concept vector from expert trajectories to score candidate actions for LLM agents, achieving top cross-benchmark averages on TextWorld, ALFWorld, and ScienceWorld.

0 favorites 0 likes
#in-context-learning

Has GPT-6-Astra Achieved Embodied Intelligence?

Hugging Face Daily Papers ↗ · 2026-10-15 Cached

该报告大规模评测了 GPT-6-Astra 在 12 项机器人基准上的具身能力,指出其已跨越具身能力门槛但尚未实现实用具身智能,并探讨了 System 2/System 1 协同与真实部署中的成本、执行时间等关键挑战。

0 favorites 0 likes
#in-context-learning

Copy-Paste Language Models

Hugging Face Daily Papers ↗ · 2026-10-13 Cached

论文提出 Copy-Paste Language Models(CPLMs),在传统语言建模头之外加入轻量 Copy-Paste 头以增强模型的上下文复制能力,在 135M 到 1.7B 参数规模的预训练中验证了训练效率、下游性能及长上下文检索的提升,且开销极小。

0 favorites 0 likes
#in-context-learning

GIST: Generator-state-Informed Supervision for Tabular Foundation Models

arXiv cs.LG ↗ · 2026-10-12 Cached

论文提出 GIST(Generator-state-Informed Supervision for Tabular Foundation Models),利用生成器中间状态的条件标签分布替代硬标签监督,以降低表格基础模型预训练中的采样方差,在 TabArena 和 TALENT 基准上取得更好的平均排名。

0 favorites 0 likes
#in-context-learning

Neural expectation-maximization and meta-models

arXiv cs.LG ↗ · 2026-10-12 Cached

该论文提出将 Transformer 网络解释为混合概率推理过程的可学习近似:自注意力对应在输入条件化、动态结构的马尔可夫随机场上执行 max-product 信念传播,而层间深度演进则对应隐式的期望最大化(EM)过程。基于此视角,作者提出 meta-model 范式,将自注意力、混合专家和门控单元等机制统一起来。

0 favorites 0 likes
#in-context-learning

An all-transistor physical-learning transformer can be trained to perform in-context learning

arXiv cs.LG ↗ · 2026-10-12 Cached

该论文提出了一种全晶体管物理学习transformer,通过在双电路中引入额外电路元件恢复双向信息流,并利用亚阈值MOSFET的指数电流定律构建注意力机制,仿真表明该电路可被训练执行简单的上下文学习任务。

0 favorites 0 likes
#in-context-learning

Relational Learning as World Modeling

Hugging Face Daily Papers ↗ · 2026-10-12 Cached

论文提出 TabWM(表格世界模型)与 WorldRel 生成器,通过在上下文中观察执行样本来推断关系型系统的动态,预测动作导致的稀疏结构化编辑,从而将关系学习从状态建模扩展到世界建模与动作规划。

0 favorites 0 likes
#in-context-learning

SIMIT: Self-Improving Vision-Language Models via Imagination at Test-Time

Hugging Face Daily Papers ↗ · 2026-10-11 Cached

SIMIT is a test-time self-improvement framework where vision-language models generate their own query-specific practice data via imagination, achieving up to +7.20% mean relative gain over BAGEL-7B across 17 benchmarks without external annotations.

0 favorites 0 likes
#in-context-learning

SFT-as-Context Mitigates Forgetting in Supervised Fine-Tuning

arXiv cs.CL ↗ · 2026-10-09 Cached

The paper proposes SFT-as-context, a two-pass inference method where the parent model uses the fine-tuned model's response as context, mitigating catastrophic forgetting of general capabilities after supervised fine-tuning while preserving task-specific improvements.

0 favorites 0 likes
#in-context-learning

Spatial Induction Heads: In-Context Learning of Multidimensional Cellular Automata

arXiv cs.LG ↗ · 2026-10-08 Cached

本文提出"空间感应头"(spatial induction heads),一种两层 gather-and-match 电路,解释 Transformer 如何在多维细胞自动机的序列化数据中进行上下文学习。研究表明,该电路可逼近贝叶斯后验并在确定性规则上实现精确预测,训练后的两层 Transformer 能泛化到未见过的规则。

0 favorites 0 likes
Next →
← Back to home

Submit Feedback