The paper introduces NEEDLE, a training‑free technique for removing backdoors from large language models. After a trigger is identified, NEEDLE estimates a backdoor direction and a refusal subspace using activation vectors, then applies sequential weight orthogonalisation to suppress the backdoor while preserving refusal‑related representations. The method requires no clean reference model or original poisoned data and achieves the lowest attack success rate and minimal impact on model performance across multiple model families and attack types.
By Minoo Kim, Vasileios Lampos, George Drayson
The article outlines a vision and recent progress toward a decentralized system that learns from real‑time building data, treating the building itself as a data center to enhance privacy and reduce dependence on external clouds. It introduces a lightweight, Kubernetes‑like orchestration framework for deploying AI services on low‑power microcontrollers, such as those in the Arduino ecosystem, enabling in‑situ learning on sensors. The work also presents experimental results for new online learning algorithms and proposes a cloud‑fog‑edge architecture using KOptim and FIWARE components.
By Christophe C\'erin, Mamadou Sow, Fr\'ed\'eric Andr\`es
The paper introduces a ground‑truth framework for disentangling uncertainty into epistemic and aleatoric components using sample‑conditional pointwise posterior risk. It evaluates current methods, finding that Spectral‑normalized Neural Gaussian Processes and Variational Latent Gaussian Processes best recover the ground‑truth uncertainty, while most methods align more closely with posterior variance and miss predictor bias. The study also explores the entanglement of estimated uncertainties and the impact of modeling choices, providing practical guidance and releasing 13 semi‑synthetic datasets for further validation.
By Frieder Wizgall, Georg Tirpitz, Moritz Seiler, Kerstin Ritter, B\'alint Mucs\'anyi
The paper argues that evaluating anonymity in synthetic data generation must focus on the generative model rather than just the resulting dataset. It interprets GDPR definitions of personal data and anonymization under realistic model-access scenarios, mapping these to state‑of‑the‑art privacy attacks. The authors conclude that synthetic data alone is insufficient for anonymization, and that Differential Privacy offers stronger protection than Similarity‑based Privacy Metrics.
By Georgi Ganev, Emiliano De Cristofaro
arXiv:2610. 00209v1 Announce Type: new Abstract: Reliable recovery of missing measurements is important for monitoring and analysis in energy time-series systems, where fine-grained measurements may contain sensitive temporal information.
By Huizhen Huang, Yu Li, Tao Huang, Chen Hou
The paper proposes a structured version of the Information Bottleneck (IB) that separates label-relevant structure from within-condition variation using a dual-bottleneck formulation. It introduces a conditional KL term that targets within-condition information, allowing explicit control over nuisance-like variation in learned representations. Experiments demonstrate improved performance in low-data classification and consistent gains on dense prediction tasks.
By Jingyao Zhang, Yuxuan Li, Lu Han, Ali Anaissi, Nguyen H. Tran
The paper reports a reproducible study of evasion attacks on image and text classifiers. A compact convolutional network on MNIST achieved 98.63% clean accuracy but dropped to 60.20% under FGSM with ε=0.15 and 1.72% with ε=0.30, while PGD reduced accuracy to 32.47% and 0.41%; a bit‑depth‑reduction defense only partially restored performance. In contrast, a DistilBERT model fine‑tuned on the SMS Spam Collection reached 98.75% accuracy and 94.96% F1‑score, yet a sequence of predefined perturbations produced only modest probability shifts and did not flip spam to ham predictions.
By Parker Hummel (Minot State University), Ryne Skabo (Minot State University), Muhammad Abusaqer (Minot State University)
The paper investigates how reducing numerical precision through model quantization impacts the vulnerability of neural networks to model inversion attacks. It provides theoretical bounds on mutual information changes and identifies data-dependent effects, especially at 4‑bit precision. Based on these findings, the authors propose a privacy‑aware post‑training quantization strategy that allocates bits adaptively, calibrates activation ranges, and jointly optimizes weight and activation scaling to improve inversion resistance while preserving model utility.
By Rongke Liu, Youwen Zhu
The study investigates whether acoustic factors encoded in pretrained self‑supervised learning (SSL) models can systematically influence predictions in speech‑based Alzheimer's disease (AD) assessment. Using the ADReSSo dataset and three large SSL backbones, the authors applied controlled noise and reverberation interventions to various audio segments and combined layer‑wise decoding, input‑ and representation‑space interventions, and geometric alignment analysis. The results demonstrate that such acoustic interventions alter AD predictions across all models, with noise producing the strongest effect, and that these effects are structured relative to the classifier’s decision direction and reproducible on a held‑out test set.
By Serli Kopar, Alkis Koudounas, Roshan P. Rane, Sam Gijsen, Paula A. Perez-Toro, Kerstin Ritter
The paper examines the reliability of membership inference attack (MIA) vulnerability evaluation. It identifies two weaknesses: finite‑sample bias from sampling shadow datasets from a fixed superset, and miscalibration when aggregating true positive rates across individuals at very low false positive rates. The authors propose simple fixes that avoid extra computational cost and suggest further improvements with additional computation.
By Joonas J\"alk\"o, Gauri Pradhan, Ossi R\"ais\"a, Antti Honkela
The paper investigates how single‑hidden‑layer MLPs can fit training data yet fail to recover the underlying rule, focusing on higher‑order interactions and nuisance inputs. Using synthetic parity tasks, the authors benchmark different optimizers (SGD, Adam, Muon) and show that while all achieve perfect accuracy on second‑order interactions, performance drops sharply for higher orders, with Muon outperforming the others at fourth order. Experiments also reveal that freezing or removing nuisance‑related weights dramatically alters training outcomes, highlighting the role of nuisance learning in shaping the rules a shallow network can represent.
By Gongyue Zhang, Honghai Liu
The paper studies Kolmogorov‑Arnold Networks (KANs), a neural architecture that treats activation functions as learnable components, offering improved interpretability for scientific applications. It investigates how KANs scale with dataset size on image classification tasks (MNIST, Fashion‑MNIST) and a magnetic‑parameter regression task, revealing a broken neural scaling law that transitions from a faster to a slower decay of test loss as data grows. The authors also analyze how the learned activation functions evolve from simple linear approximations to more complex, interpretable symbolic forms as more data is provided.
By Tilen Cadez, Sanghoon Lee, Kyoung-Min Kim
The paper introduces a latent information sharing scheme for federated learning that mitigates client drift by sharing a small amount of hidden‑layer activations. The authors demonstrate both theoretically and empirically that this approach improves training efficiency while maintaining convergence guarantees and data privacy. Compared to existing methods such as FedProx, SCAFFOLD, FedPVR, FedProto, and SplitFed, the proposed method achieves higher model accuracy within a fixed round budget without adding significant communication overhead.
By Seungjun Lee, Ensieh Khazaei, Dimitrios Hatzinakos, Baturalp Buyukates, Sunwoo Lee
The paper develops a finite‑measurement framework for inferring the symmetry group that a quantum learning model should respect, based on candidate transformations and limited data. It shows that observable‑invisible transformations correspond to the stabilizer of a projected state when the probe span is invariant, and that recovered generators form a valid subgroup with a continuous invisible space identified via its Lie algebra. The authors introduce an unbiased shadow statistic that improves estimation rates, establish optimal gap dependence through a commuting‑qubit lower bound, and provide tools for task validation, bias quantification, and capacity analysis, all illustrated with Ising‑chain calculations.
By Zeyu Chen
The paper extends mechanistic interpretability of large language models by modeling concepts as low‑dimensional non‑linear manifolds rather than linear subspaces. It introduces a concept‑based alignment (CBA) score to compare these manifolds across layers and models, revealing block structures in intermediate layers, a shift from syntax‑dominated to mixed syntactic‑semantic concepts, and training‑dependent multilingual sharing. The study also shows that alignment patterns differ across model families and training stages, with adjacent stages aligning more closely than distant ones.
By Tido Specht, Elias Benedict Krey, Nils Neukirch, Nils Strodthoff
The paper introduces a new framework for monitoring the Pantograph‑Catenary System (PCS) that localizes PCS height and stagger by aligning video‑derived measurements with nominal GPS coordinates of the reference route. It also presents a collective anomaly detection method to assess PCS health conditions. The authors evaluate both localization and detection on a real‑world dataset from an Italian railway company, covering multiple train journeys.
By Francesco Vitale, Hangli Ge, Francesco Flammini
The study analyzes 22 OULAD asynchronous online courses (over 22,000 learners) using Zigzag Persistent Homology to track the number of disconnected behavioral clusters, denoted $eta_0$. It finds that changes in $eta_0$ strongly co‑vary with active learner counts, indicating that $eta_0$ is a participation‑sensitive indicator rather than a cause of dropout. Assessment deadlines trigger fragmentation in 82.6% of cases and the full Fragment First, Converge Later cycle in 60.2%, with long‑term fragmentation dominating 90.9% of courses.
By Hitoshi Inoue, Koichi Yasutake
Large language models (LLMs) are widely used as automatic judges, with validity typically assessed via alignment with human scores. However, aggregate agreement fails to reveal whether humans and LLMs...
The paper introduces a compact framework that transforms continuous multimodal workplace video into a structured Procedural State Memory called a Work Environment Model (WEM). Using event segmentation theory, it detects segment boundaries based on changes in visual context, location, motion, narration, gaze/object interaction, and optional exocentric workspace evidence, then abstracts each segment into an evidence‑linked event card. These event cards incrementally update the WEM, enabling efficient, auditable documentation and retrieval while respecting on‑premise privacy constraints, and the authors evaluate the system on segmentation quality, memory compression, retrieval fidelity, and long‑horizon QA.
By Vivek Chavan, J\"org Kr\"uger
The paper introduces the Distributionally Robust Schr"odinger Bridge (DRSB), a method that learns a single controller capable of handling uncertainty in the initial distribution for stochastic transport tasks. DRSB’s objective combines control energy with a KL penalty on the terminal distribution, and it seeks to minimize the worst‑case value of this objective over an ambiguity set around the nominal initial distribution. The authors derive a variational formulation, connect it to stochastic optimal control and distributionally robust optimization, and propose an alternating algorithm with Wasserstein and Sinkhorn variants. Experiments on two‑dimensional transport and image‑to‑image translation demonstrate improved robustness to input perturbations compared to standard SB, while also achieving lower mean sliced Wasserstein distance on Gaussian mixture transport.
By Jinhwan Sul, Panagiotis Theodoropoulos, Vincent Pacelli, Jaemoo Choi, Evangelos Theodorou