The paper evaluates Variational Autoencoders, Generative Adversarial Networks, and Diffusion Models for unsupervised anomaly detection in federated predictive maintenance. It compares full and partial federation setups, showing that full federation improves GAN performance but VAE and DDPM models are more robust. Decoding‑only sharing in Diffusion Models achieves similar utility to full federation while cutting communication costs.
By Usevalad Milasheuski, Piero Baraldi, Enrico Zio, Stefano Savazzi
TextCloak is a reinforcement‑learning driven framework that generates unlearnable examples (UEs) to protect textual data from unauthorized fine‑tuning of large language models (LLMs). It transforms clean text into perturbations that preserve meaning and naturalness while degrading the utility of models trained on them. Experiments on six datasets and nine LLMs show that TextCloak consistently impairs unauthorized fine‑tuning and remains robust across different architectures, training settings, and adaptive attacks.
By Chengshuai Zhao, Pingchuan Ma, Dawei Li, Bohan Jiang, Zhiyuan Yu, Zhen Tan, Huan Liu
The paper introduces Actionable Case-Based Feature Importance (A-CBFI), a method that transforms the descriptive decomposition of Localized-CBFI into a prescriptive, interventional diagnosis using a structural causal model. A-CBFI selects the cheapest single action when possible and only considers joint actions if a single action cannot achieve the desired outcome. Experiments on ten tabular benchmarks show that A-CBFI matches or exceeds the success of untargeted causal beam search while prescribing fewer direct actions and requiring fewer model evaluations.
By Sejong Oh
The paper introduces a system that enables first‑order LoRA training on phone NPUs by running all frozen‑weight multiplications on the NPU while keeping adapters and the optimizer on the CPU. It reuses weight tiles for input‑gradient computation, selects precision based on gradient sensitivity, and fuses remaining CPU operations to overlap with NPU execution. The approach is evaluated on Qualcomm Snapdragon devices with Qwen and Llama LLMs, demonstrating background training for nightly personalization and benchmark workloads.
By Xu Zheng, Wei Cheng, Zhuomin Chen, Dongkuan Xu, Mo Sha, Jingchao Ni, Dongsheng Luo
The paper introduces PAI-KNN, a k‑nearest neighbor classifier designed for data stream classification with concept drift and class imbalance. Unlike traditional methods that modify memory through instance removal or resampling, PAI‑KNN keeps a simple sliding‑window buffer and adapts its decision rule at inference time by adjusting class priors based on recent observations. Experiments on synthetic streams with varying imbalance and drift show that PAI‑KNN performs competitively with state‑of‑the‑art techniques, especially when dynamic class imbalance and concept drift are present.
By Tobias Callies (University of the Bundeswehr Munich), Eirini Ntoutsi (University of the Bundeswehr Munich), Arthur Zimek (University of Southern Denmark)
The paper introduces the Multivariable Predictability Score (MAPS), a model‑agnostic metric designed to quantify the intrinsic predictability of multivariable time‑series forecasting tasks. MAPS leverages covariate information across various lookback windows and forecasting horizons to assess predictive information, addressing limitations of existing measures that often miss nonlinear dependencies or focus on univariate series. Experiments on synthetic and real‑world datasets show MAPS aligns closely with expected reference predictability (r = 0.963) and provides more reliable projections than six baseline measures.
By Yuxuan Shu, Vasileios Lampos
The paper introduces a new framework for activation steering in large language models, distinguishing between the steering direction (estimand) and the method used to recover it (estimator). It shows that most existing techniques can be viewed as plug‑in estimators of the average gradient of an outcome regression, and proposes a complementary estimator based on the Stein score to directly target the gradient. By combining these approaches, the authors achieve an orthogonalized estimator with double‑robustness properties and demonstrate its effectiveness on benchmark datasets.
By David Arbour, Erfan Jahanparast, Serina Chang, Avi Feller
arXiv:2610. 33259v1 Announce Type: new Abstract: Classical optimization and learning-guided search offer complementary approaches to open problems in discrete mathematics, but their outputs are difficult to compare when search spaces, validity checks, and stopping rules differ.
By Xiaoan Xu, Ethan X. Fang, Jesus A. De Loera, Junwei Lu, Luze Xu
The paper introduces the Clustering Occupation Kernel (CLOCK) method, a representation-based framework that embeds time series from dynamical systems into a reproducing kernel Hilbert space using a time‑averaged occupation kernel. This embedding allows trajectory comparison via the RKHS norm and, for differential equation systems, approximates the kernel mean embedding of the invariant Sinai‑Ruelle‑Bowen measure, effectively measuring the maximum mean discrepancy between attractors. CLOCK handles unequal‑length and irregularly‑spaced data, incorporates Takens’ embedding for general time series, and demonstrates superior performance over benchmark methods on synthetic data and real intracranial EEG recordings, with ablation studies confirming theoretical convergence and noise robustness.
By Michael L. Wells, Kamel Lahouel, Victor Rielly, Bruno Jedynak
Geodesic‑SVDD introduces a distributed learning framework for structured tabular data that transforms each client’s task into a one‑class density estimation on a hypersphere, enabling aggregation of local models as probabilistic manifolds in a single communication round. The method employs a Residual Spherical Encoder and a Geodesic Latent Mixup to enforce strict hyperspherical constraints, preventing manifold collapse, and proposes a One‑Shot Product‑of‑Experts protocol for privacy‑preserving aggregation without iterative synchronization. Experiments on five tabular benchmarks and varying label skew levels show that Geodesic‑SVDD achieves the highest mean AUROC among seven federated methods, surpasses federated tree ensembles by up to 21 AUROC points, and its One‑Shot variant matches centralized Deep SVDD under IID and mild skew conditions.
By Andrea Moleri, Christian Intern\`o, Markus Olhofer, Barbara Hammer, Ali Raza
The paper introduces the Min-Marginal Power Index (MinPI), a game-theoretic feature selection method that labels features as harmful, neutral, or decisive based on whether adding them to a coalition drops utility below a threshold. MinPI enables conservative feature removal without a score cutoff, and its application to six benchmark datasets removed 12.5% of features with negligible loss in held‑out utility. The method also improves multi‑objective search performance by prefiltering features, especially in datasets with many removable features.
By Wenjie Shi, Chi Zhao
The paper investigates how non‑monotonic performance dynamics in single‑task reinforcement learning affect policy selection. By introducing post‑hoc diagnostic metrics, the authors evaluate common selection strategies on continuous‑control benchmarks and find that typical practices often miss the best policy, leading to suboptimal performance and wasted computation. They propose a validation‑based selection method that better approximates the best visited policy while improving efficiency.
By Judith Echevarrieta, Etor Arza, Aritz P\'erez, Josu Ceberio
The paper introduces an error‑diffusion (ED) optimizer that replaces the full‑precision latent copy of binary‑weight language models with a small 2‑8‑bit residual buffer, reducing training state per weight from 97 bits to 3‑9 bits—up to 19× less. ED accumulates unrealized weight updates in the buffer and flips the binary weight when the accumulated value crosses a threshold aligned with the quantization grid, enabling efficient training on large transformers with minimal loss compared to latent‑weight baselines. The method scales to models with up to 19 M parameters on enwik8 and 70 M on wikitext‑103, maintaining performance within a few percent of reference while allowing on‑device adaptation within a 275 MB memory budget.
By Pierre-Olivier Bonin
Sparse Reconstruction by Direct Measurement-to-Field Generation in Function Space introduces a functional rectified flow framework called Direct Measurement-to-Field Generation (DMF) that learns to generate physical fields directly from sparse measurements. Unlike traditional grid-based methods, DMF jointly learns a sensor–global–sensor representation and a generative transport mechanism, enabling dynamic fusion of local grounding with global context to evaluate fields at arbitrary query coordinates. Benchmarks demonstrate that DMF outperforms baselines in scenarios with unknown boundaries, missing high-resolution training data, and unmeasured quantities, while its stochastic ensemble generation provides reliable uncertainty quantification and improved continuous ranked probability scores.
By Linzheng Wang, Jason Chen, Nicolas Tricard, Zituo Chen, Xingsen Guo, Sili Deng
MINT is a graph neural network framework designed to detect illicit accounts in financial transaction networks. It addresses structural inconsistency by using an attentive message‑passing scheme that tracks deviations between nodes in bi‑directional flows and a transaction salience module that regularizes node interactions. To cope with limited labeled data, MINT employs Positive‑Unlabeled learning, leveraging a small set of known illicit nodes and many unlabeled ones, and demonstrates superior performance on three benchmark datasets.
By Yin-Ju Wu, Bi-Cheng Yan, Berlin Chen
The paper introduces localized case-based feature importance (CBFI), a framework that decomposes individual predictions in complex tabular models into distinct structural components, separating independent main effects from pure interactions. Unlike additive explanations such as SHAP and LIME, CBFI explicitly captures feature synergies, enabling the recovery of latent interaction structures. Experiments on wine quality classification and medical insurance regression, along with fidelity tests and synthetic benchmarks, show that CBFI accurately disentangles interactions hidden by additive models and aligns with the nonlinear dependencies in gradient boosting and deep neural networks.
By Sejong Oh
The paper introduces STeP-TS, a compact one‑layer convolutional model for multivariate time‑series forecasting that learns a shared bank of multi‑scale temporal filters for each variable. Lightweight variable‑specific scalar weights modulate the contribution of each filter, allowing the model to capture both shared temporal patterns and variable‑specific dynamics. The learned weights and filter‑output energy are used for structured pruning, reducing computational complexity with minimal loss in performance, and an amplitude‑adaptive normalization preserves variable‑dependent temporal amplitudes.
By Hanne Raum, Jasper Hoffmann, Joschka Boedecker
The paper introduces FlashSO(2), a framework that uses tailored low‑precision emulation under the integer Ozaki scheme to accelerate SO(2)-equivariant layers in graph neural networks for materials modeling. FlashSO(2) outperforms native low‑precision formats such as TF32 in throughput, accuracy, and memory usage, achieving up to 40× speed‑ups on NVIDIA Hopper GPUs and enabling faster training and simulation of ML interatomic potentials and Hamiltonian models.
By Denghui Lu, Alexander Maeder, Alexandros Nikolaos Ziogas, Mathieu Luisier, Manasa Kaniselvan
The paper introduces Conversational Generalized Linear Bandits with Hybrid Feedback, a framework that blends absolute item rewards (such as clicks or ratings) with relative key‑term comparisons. It proposes the Conversational UCB with Hybrid Feedback (HyConUCB) algorithm, which builds unified confidence bounds from heterogeneous feedback and adaptively selects key‑term comparisons to balance information gain and directional coverage. A refined regret analysis shows that HyConUCB outperforms reward‑only and dueling‑only baselines, and experiments on synthetic and real‑world data confirm its effectiveness.
By Yurui Zou, Qirun Zeng, Jinhang Zuo
FOCUS is a reinforcement learning method designed to improve long‑horizon language‑model agents by using selective branching to train a critic network more accurately. The rollout controller actively samples continuations from states that are expected to reduce the critic’s estimation error, and the refined value estimates are used to score segments between consecutive states with a TD(λ) advantage. Experiments on legal reasoning, deep search, and code localization tasks show that FOCUS outperforms baselines such as GRPO and VinePPO, achieving significant accuracy gains on benchmarks like BrowseComp‑Plus and DeonticBench.
By Zhengping Jiang, William Jurayj, Liaoyaqi Wang, Chunsheng Zuo, Anqi Liu, Benjamin Van Durme