Tag
The paper introduces the first active learning method for contextual linear optimization, using a margin-based label acquisition rule guided by the Smart Predict-then-Optimize (SPO) loss that requires fewer labels than supervised learning to achieve the same excess risk, with demonstrations in personalized pricing and shortest-path problems.
This paper derives a risk-optimal active learning acquisition criterion when a labeling oracle may abstain and its response probability shares parameters with the classifier, using Fisher information adjusted for nuisance parameters combined with local curvature of multiclass zero-one risk. Simulations and a Consumer Complaints replay show response-aware acquisition outperforms response-discounted baselines at larger query budgets.
The paper introduces Bayesian κ-Explorer, which places Dirichlet priors on MDP transition distributions to simplify minimax transition estimation by replacing confidence-set optimization with a closed-form posterior variance exploration signal, while retaining the Õ(T^-1/3) convergence rate of the frequentist κ-Explorer and showing lower worst-case estimation errors on Pendulum, Mountain Car, and Taxi.
This paper proposes a Bayesian framework for interactive program induction where a language model serves as a prior over programs, deriving a greedy query-selection algorithm that provably converges after O((1/γ) log 1/p⁰(f*)) informative oracle queries. On the CodeARC benchmark, it improves accuracy over LM-only agents by up to 22 percentage points while using up to 10.5 fewer input-output examples.
The paper presents a method for learning word-level explanations of text reuse in Latin literature with minimal annotation, using synthetic training pairs plus active learning; a joint pointer network matches full-data performance with only 200 annotated pairs out of 1,490 references.
The paper characterizes when unlabeled data pools can substitute for membership queries in learning halfspaces, showing near-optimal label complexity under mild angular assumptions on the marginal and separating results when those assumptions fail.
The paper introduces Conditional Poisson Sampling as a fixed-size drop-in replacement for Poisson Sampling in active statistical inference, showing it achieves no worse and often better asymptotic variance while eliminating random sample size issues, along with a released Python package.
该论文提出层级分数后验恢复方法(HSPR),通过结合历史响应模式与自适应任务选择,仅用小子集即可高效估计软件代理在完整基准测试上的分数,在SWE-bench系列任务库上取得最低的平均与最大绝对误差。
该论文提出 acquisition-time residual adaptation 方法,在固定实验预算下保留可复用的基础打分器并学习排名修正,使 20 个测试筛选任务的命中回收率从 22.58% 提升至 28.61%(相对提升 26.7%),并优于多种对照方法。
该论文提出福利感知偏好获取方法(WAPA),在有限人工反馈预算下,通过社会 welfare 目标来分配人工标注,以改善多元对齐中不同个体偏好的监督质量,实验证明仅获取 20% 人工标签即可恢复 75–86% 的效用差距。
SynAct提出一种主动学习框架,通过在重复采样中计算决策与最终结果之间的互信息,将稀缺的专家反馈精准分配到最影响结果的决策点上,并将修正蒸馏为可复用规则,在四个基准、三个模型家族的全部16个设置中平均提升11.7个百分点。
Moonworks 推出 Lunara 论文,提出一种参数量低于 10B 激活参数的 Diffusion Mixture Transformer 架构,用于图像生成中的艺术智能建模,并在 GPT-5.6 Sol 评测及人工盲评中取得最佳审美质量得分。
Microsoft researchers propose ActiveSaddler, a method that selects training tasks for agent harness auto-tuners based on unresolved failure patterns, boosting pass rates by 4.4 points on GAIA2 and 7.5 on Terminal-Bench 2.0 while cutting tuning costs sharply.
Dyna Robotics addresses the reliability gap in robotics by combining a large-scale data pyramid, a specialized reward model, and iterative deployment to achieve near-perfect task performance, moving towards commercial-grade autonomy.
This review article synthesizes active learning research for biodiversity monitoring, addressing label efficiency and highlighting the need for methods that support validation and reliable ecological inference.
This paper introduces ActiveContinuous-Score Occupancy Modeling to optimize the allocation of review efforts for machine learning labels in AI-assisted biodiversity surveys, enhancing the accuracy of occupancy models.
This paper introduces Helix-FNO, a teacher-student architecture that couples a high-fidelity mechanistic model with a Fourier neural operator to learn solution operators for fast surrogate simulation in process engineering, enabling millisecond inference times.
This paper introduces HACK GPs, a method that uses online learning with expert advice for kernel selection in Gaussian Processes, enhancing robustness in sequential decision-making tasks such as Bayesian optimization and active learning.
This paper proposes a Bayesian decision formulation for multiobjective optimization under uncertainties, emphasizing expected hypervolume maximization with Gaussian Processes and active learning strategies.
This technical note presents a decision-focused active learning approach for optimizing critical materials recovery, showing that adaptive policies can achieve maximum enrichment with fewer experiments than nonadaptive methods in autonomous laboratory workflows.