Tag
FrontierMat introduces a benchmark of 330 validated frontier materials R&D problems for language models, finding the strongest model scores only 53.7% and that a domain-adaptation method lifts an open-weight model to 55.4%, matching proprietary leaders.
论文提出 Bottle-SPU（Bottle-SKU）大规模多SKU洗发水瓶表面缺陷检测基准，包含13,119张图像与20,419个实例级标注框，并提出原型引导的动态域调制方法 ProDMM，通过SKU视觉原型生成低秩Query/Value调制以提升跨产品变体的缺陷检测泛化能力。
Researchers from Los Alamos National Laboratory and NVIDIA present an expert-curated benchmark of 250 questions on inertial confinement fusion to evaluate LLMs' scientific fluency, physical reasoning, and quantitative analysis. They compare frontier open/closed models with Prospero, a domain-specialized model, finding that specialization improves scores but critical physics errors persist, and recommend assessing existing capabilities and retrieval benefits before investing in further specialization.
提出了一个跨域迁移表示特征的统计框架，用于检测 LLM 的幻觉与推理错误，在新域仅需 25 条标注样本即可将幻觉检测 AUROC 提升 6.8%、推理步骤预测 AUROC 提升 8.4%。
论文提出 Skill Schema Transport (SST) 框架，通过将新领域知识重铸为模型已有技能空间中的可复用程序化 schema 来缓解持续领域适应中的灾难性遗忘，在两个 3B 模型上将 Qwen MATH 流的最大遗忘从 42.41 降至 7.10 个点。
CorteXforge is a scenario-based framework that generates realistic, reproducible annotated RF datasets on programmable SDR testbeds (SLICES-RI/CorteXlab), validated via automatic modulation classification showing synthetic-to-real gaps and the benefits of synthetic pretraining.
论文提出 WebMix，一种利用混合超网络在部署时组合已有低秩适配器（LoRA）的网页智能体适配方法，无需在目标界面上重新训练即可实现跨域和跨版本迁移，并在 TimeWarp 与 WebArena-Lite-v2 上取得显著提升。
AgriJev 将 Qwen3.5 系列（0.8B–9B）适配为农业领域的候选评分模型，通过单次前向传播直接输出候选分数与相对概率；AgriJev-9B 在 17 个外部测试视图上达到 75.44% 平均准确率，且评分延迟约为同规模答案生成的一半。
GlucoPRISM 通过协议监督的预训练，将连续血糖监测（CGM）日表征分解为 Trait、State 和 Sensor 三个频带，在部署时以固定投影移除传感器相关分量，从而在无需目标数据的情况下将十二个跨队列迁移方向的平均 AUROC 从 68.75 提升至 73.57。
本文提出一个基于大语言模型的框架，利用微调的开源 LLM 自动化生成交通管理计划（TMP），并在威斯康星州交通部的 WisTMP 系统中进行案例研究，发现微调显著提升性能，但模型仍难以生成项目特定理由和准确的成本估算。
本文研究流式说话人分离模型（Streaming Sortformer v2）在小数据适配时出现的'看似提升实则遗忘'现象，揭示聚合指标（DER）掩盖的身份一致性退化，并探讨rehearsal在恢复与跨域迁移之间的权衡。
This paper proposes BaryFM, a flow matching model that learns to approximate the entire Wasserstein simplex — the full family of Wasserstein barycenters over the simplex of weight vectors — rather than a fixed-weight barycenter. BaryFM achieves the best average rank among 15 methods across 10 domain adaptation benchmarks and is validated on tasks including generalization, Bayesian posterior aggregation, and algorithmic fairness.
The paper introduces CALIBRA, a calibration-first multimodal temporal framework for cross-cohort asthma risk forecasting that uses recurrent encoders, gradient-reversal domain adaptation, hierarchical logistic calibration, and conformal prediction to remain reliable under distribution shift and incomplete modalities, validated on a semi-synthetic three-cohort benchmark.
This paper studies how to specialize large language models for generating executable algorithmic trading code, showing that framework-oriented continued pretraining plus supervised fine-tuning on validated request-to-code pairs greatly improves success on the QuantCode-Bench benchmark and agentic strategy generation, while also revealing a capability-retention failure where domain specialization degrades structured tool calling.
This paper introduces a preference-guided adaptation framework for open-vocabulary semantic segmentation that leverages prompt disagreement to generate binary preferences, eliminating the need for dense pixel-level annotations and improving performance in specialized domains.
This paper introduces TAM-Chain, a multi-scale thyroid cytology classification method that employs absorbing Markov chains and Shannon entropy for uncertainty quantification, aiming to suppress false negatives and adapt to domain shifts.
The paper proposes EviGDA, a framework that enhances graph domain adaptation by combining graph-aware and graph-free experts to improve prediction under structural shifts.
The paper proposes TSS, a target-side sparsification framework for speculative decoding in domain-specific LLMs, which improves inference efficiency and performance by skipping selected target layers.
This academic paper presents methods for tutoring large language models to improve their domain adaptability, precision, and safety.
This paper evaluates intrinsic sequence-likelihood confidence in retrieval-dominated extractive QA, finding that confidence signals are insufficient for model control and that retrieval alone achieves high accuracy under pre-specified criteria.