machine-learning

本文提出将球员的最短可行到达时间（Time-to-Arrival）建模为条件分位数回归问题，并在2022世界杯PFF光学追踪数据上训练的轻量MLP模型在多项指标上超越了六种基于物理的球员运动模型，且展示了跨运动项目（篮球、NFL）的迁移能力。

A multi-site benchmark of empirical, machine-learning (LightGBM) and deep-learning (U-Net) models for satellite-derived bathymetry in the Arabian Gulf, validated leakage-free against 285,000 multibeam soundings, showing that validation design matters more than model choice, with the simple Stumpf ratio outperforming complex models in transfer settings.

论文提出 CD-FFM，一种基于似然函数流匹配（likelihood functional flow matching）的生成式气候降尺度模型，可将再分析数据降尺度至1公里分辨率，在芬诺斯坎迪亚日气温数据上优于回归、神经算子和扩散模型，且采样速度极快并能给出精确似然。

本文提出 FG-ConvNO（Factorized Geometry Convolutional Neural Operator），一种面向几何感知的神经算子，用于在有限训练数据下对螺旋桨表面压力和总推力进行高精度 CFD 预测，并结合 DISCO 块实现离散化无关建模。

The paper introduces the Clustering Occupation Kernel (CLOCK) method, a representation-based framework that embeds time series from dynamical systems into an RKHS to enable clustering via maximum mean discrepancy between attractors, validated on synthetic data and intracranial EEG recordings.

该论文证明了差分隐私算法在组合与子采样下的隐私损失分布可通过截断后的高斯近似来刻画，从而为DP-SGD等场景给出比Rényi分析更紧的闭式隐私界，尤其在采样率1/t时显著降低所需噪声规模σ的要求。

该论文提出局部预测一致性（local predictive consistency）作为成对个体公平性定义的节点级松弛，通过相似度加权的邻域平均预测来生成可解释的公平性分数，并将其扩展到二分类、回归和生存分析场景。

A UCL paper proposes MAPS, a model-agnostic metric for quantifying the intrinsic predictability of multivariable time series forecasting tasks, evaluated against 66 baseline measures on synthetic and real-world datasets with strong correlation to reference predictability (r=0.963).

提出 PAI-KNN，一种通过推理时先验调整而非内存修改来适应概念漂移与类不平衡的 KNN 数据流分类器，在合成数据流基准上取得了与最先进方法竞争的性能。

Voxel51 通过 Product Hunt 推出其多模态数据平台，帮助机器人团队从视频和传感器数据中发现失败案例、理解原因并补齐训练数据，从而加速物理 AI 模型的迭代。

提出在分布漂移下进行在线模型测试的方法学，通过区分人口变化与数据完整性问题两种漂移机制，在受控基线上复现漂移以估计模型当前性能，并评估修复数据完整性问题的潜在收益。

A new paper introduces LPDS, a generative inverse modeling framework that maps desired fusion outcomes and target configurations to optimized laser pulse shapes for inertial confinement fusion, using multi-objective physics-constrained training and control mechanisms like constraint-conditioning and inpainting.

This paper presents an AI-augmented decision-support system for ambulance travel time estimation using historical Skåne region data, showing that travel times vary by hour, day, and trip priority while leveraging SHAP and permutation importance for interpretability.

提出 PFDF，一种基于路径依赖的联邦遗忘客户端选择框架，通过 CLP 加权历史记录筛选出与目标客户端弱耦合但全局信息量大的保留客户端，以更少的通信和计算成本实现高效遗忘。

MLE-World: a world model finetuned from Qwen3.5-4B that predicts ML experiment outcomes (error traces, logs, validation scores) without execution, enabling affordable test-time scaling for AI research agents. It matches real terminal outputs on 77.9% of held-out programs and boosts mean performance by up to 22.1% on MLE-bench Lite tasks.

Opinion piece arguing that rather than being overhyped, AI progress over the past 18 months has been consistently underestimated by most observers and experts.

本文评估了五个天气基础模型作为冻结编码器在增强岩石风化（ERW）碳去除量化中的迁移能力，其中基于 Aurora 1.5 构建的 TerraNova-ERW 在未见农田上表现最佳，支持少样本预测并可显著减少土壤采样需求。

The paper proposes WaferHDC, an online open-world wafer defect discovery framework based on hyperdimensional computing that dynamically detects and groups recurring defect patterns without labels or retraining, outperforming the strongest online baseline by 13.3 p.p. in adjusted Rand index on the WM-811K dataset.

This arXiv paper explores realistic models of change through a causal perspective, connecting decision-making frameworks with dynamical systems to better capture how systems evolve over time.

该论文提出一种机器学习优化的光学模型，用于分析偏振共振软 X 射线散射（RSoXS），从而以亚单层深度分辨率定量解析埋藏界面上分子的取向、化学身份与堆积状态，并发现分子取向从界面处开始的梯度演化，推翻了衬底图案化的相关假说。