Tootfinder

Opt-in global Mastodon full text search. Join the index!

No exact results. Similar results found.
@arXiv_csLG_bot@mastoxiv.page
2025-10-08 10:43:39

(Token-Level) \textbf{InfoRMIA}: Stronger Membership Inference and Memorization Assessment for LLMs
Jiashu Tao, Reza Shokri
arxiv.org/abs/2510.05582

(Token-Level) \textbf{InfoRMIA}: Stronger Membership Inference and Memorization Assessment for LLMs
Machine learning models are known to leak sensitive information, as they inevitably memorize (parts of) their training data. More alarmingly, large language models (LLMs) are now trained on nearly all available data, which amplifies the magnitude of information leakage and raises serious privacy risks. Hence, it is more crucial than ever to quantify privacy risk before the release of LLMs. The standard method to quantify privacy is via membership inference attacks, where the state-of-the-art ap…

@arXiv_csLG_bot@mastoxiv.page
2025-10-06 10:23:19

Learning Robust Diffusion Models from Imprecise Supervision
Dong-Dong Wu, Jiacheng Cui, Wei Wang, Zhiqiang She, Masashi Sugiyama
arxiv.org/abs/2510.03016 …

Learning Robust Diffusion Models from Imprecise Supervision
Conditional diffusion models have achieved remarkable success in various generative tasks recently, but their training typically relies on large-scale datasets that inevitably contain imprecise information in conditional inputs. Such supervision, often stemming from noisy, ambiguous, or incomplete labels, will cause condition mismatch and degrade generation quality. To address this challenge, we propose DMIS, a unified framework for training robust Diffusion Models from Imprecise Supervision, w…

@arXiv_csCL_bot@mastoxiv.page
2025-09-25 10:42:02

From Input Perception to Predictive Insight: Modeling Model Blind Spots Before They Become Errors
Maggie Mi, Aline Villavicencio, Nafise Sadat Moosavi
arxiv.org/abs/2509.20065

From Input Perception to Predictive Insight: Modeling Model Blind Spots Before They Become Errors
Language models often struggle with idiomatic, figurative, or context-sensitive inputs, not because they produce flawed outputs, but because they misinterpret the input from the outset. We propose an input-only method for anticipating such failures using token-level likelihood features inspired by surprisal and the Uniform Information Density hypothesis. These features capture localized uncertainty in input comprehension and outperform standard baselines across five linguistically challenging d…

@arXiv_csCR_bot@mastoxiv.page
2025-10-07 08:03:49

CryptOracle: A Modular Framework to Characterize Fully Homomorphic Encryption
Cory Brynds, Parker McLeod, Lauren Caccamise, Asmita Pal, Dewan Saiham, Sazadur Rahman, Joshua San Miguel, Di Wu
arxiv.org/abs/2510.03565

CryptOracle: A Modular Framework to Characterize Fully Homomorphic Encryption
Privacy-preserving machine learning has become an important long-term pursuit in this era of artificial intelligence (AI). Fully Homomorphic Encryption (FHE) is a uniquely promising solution, offering provable privacy and security guarantees. Unfortunately, computational cost is impeding its mass adoption. Modern solutions are up to six orders of magnitude slower than plaintext execution. Understanding and reducing this overhead is essential to the advancement of FHE, particularly as the underl…

@arXiv_eessIV_bot@mastoxiv.page
2025-10-02 08:09:40

Deep Learning Approaches with Explainable AI for Differentiating Alzheimer Disease and Mild Cognitive Impairment
Fahad Mostafa, Kannon Hossain, Hafiz Khan
arxiv.org/abs/2510.00048

Deep Learning Approaches with Explainable AI for Differentiating Alzheimer Disease and Mild Cognitive Impairment
Early and accurate diagnosis of Alzheimer Disease is critical for effective clinical intervention, particularly in distinguishing it from Mild Cognitive Impairment, a prodromal stage marked by subtle structural changes. In this study, we propose a hybrid deep learning ensemble framework for Alzheimer Disease classification using structural magnetic resonance imaging. Gray and white matter slices are used as inputs to three pretrained convolutional neural networks such as ResNet50, NASNet, and M…

@arXiv_csLG_bot@mastoxiv.page
2025-10-06 10:25:29

Signature-Informed Transformer for Asset Allocation
Yoontae Hwang, Stefan Zohren
arxiv.org/abs/2510.03129 arxiv.org/pdf/2510.03129

Signature-Informed Transformer for Asset Allocation
Robust asset allocation is a key challenge in quantitative finance, where deep-learning forecasters often fail due to objective mismatch and error amplification. We introduce the Signature-Informed Transformer (SIT), a novel framework that learns end-to-end allocation policies by directly optimizing a risk-aware financial objective. SIT's core innovations include path signatures for a rich geometric representation of asset dynamics and a signature-augmented attention mechanism embedding financi…

@arXiv_csCR_bot@mastoxiv.page
2025-10-01 09:39:08

Exact Bias of Linear TRNG Correctors - Spectral Approach
Maciej Skorski, Francisco-Javier Soto, Onur G\"unl\"u
arxiv.org/abs/2509.26393

Exact Bias of Linear TRNG Correctors - Spectral Approach
Using Fourier analysis, this paper establishes exact security bounds for linear extractors in True Random Number Generators (TRNGs). We provide the first near-optimal total variation security characterization by interpolating between optimal $\ell_{\infty}$ and $\ell_2$ norm results, expressed through code weight enumerators and input bias parameters. Our bounds improve security assessments by an order of magnitude over previous approximations. By scanning ~20,000 codes, we reveal fundamental…

@buercher@tooting.ch
2025-12-04 17:31:25

Dans un arrêt rendu le 19 novembre, le TAF vient d’en faire la démonstration en prenant position sur un recours contre la loi fédérale sur le renseignement (LRens), qui autorise de facto le SRC Š espionner massivement les Suisses, sans limite ni contrôle. Afin de respecter les droits fondamentaux ainsi bafoués, le parlement fédéral est invité, dans un délai de cinq ans, Š réviser sa copie, approuvée Š 65,5% par la population lors d’un référendum en 2017.
letemps.ch/suisse/autorisant-l