Membership inference (MI) attacks try to determine if a data sample was used to train a machine learning model. For foundation models trained on unknown Web data, MI attacks can be used to detect copyrighted training materials, measure test set contamination, or audit machine unlearning. Unfortunately, we find that evaluations of MI attacks for foundation models are flawed, because they sample members and non-members from different distributions. For 8 published MI evaluation datasets, we show that blind attacks -- that distinguish the member and non-member distributions without looking at any trained model -- outperform state-of-the-art MI attacks. Existing evaluations thus tell us nothing about membership leakage of a foundation model's training data.

News recommendation is crucial for facilitating individuals' access to articles, particularly amid the increasingly digital landscape of news consumption. Consequently, extensive research is dedicated to News Recommender Systems (NRS) with increasingly sophisticated algorithms. Despite this sustained scholarly inquiry, there exists a notable research gap regarding the potential synergy achievable by amalgamating these algorithms to yield superior outcomes. This paper endeavours to address this gap by demonstrating how ensemble methods can be used to combine many diverse state-of-the-art algorithms to achieve superior results on the Microsoft News dataset (MIND). Additionally, we identify scenarios where ensemble methods fail to improve results and offer explanations for this occurrence. Our findings demonstrate that a combination of NRS algorithms can outperform individual algorithms, provided that the base learners are sufficiently diverse, with improvements of up to 5\% observed for an ensemble consisting of a content-based BERT approach and the collaborative filtering LSTUR algorithm. Additionally, our results demonstrate the absence of any improvement when combining insufficiently distinct methods. These findings provide insight into successful approaches of ensemble methods in NRS and advocates for the development of better systems through appropriate ensemble solutions.

When users stand to gain from certain predictive outcomes, they are prone to act strategically to In this work we study classification of strategic agents that obtain predictions that are favorable. Most current choose whether to apply or not in response to the learned works consider strategic behavior that manifests classifier. Strategic candidates apply only if the expected as users modifying their features; instead, we utility from passing screening outweighs associated costs; study a novel setting in which users decide thus, application choices derive from beliefs regarding classification whether to even participate (or not), this in response outcomes. Since these choices in aggregate determine to the learned classifier. Considering learning the test-time distribution, learning becomes susceptible approaches of increasing strategic awareness, to self-selection--namely selection that is carried out by the we investigate the effects of user self-selection agents which predictions target. Our goal in this paper is to on learning, and the implications of learning on study learning under such self-selective behavior, which we the composition of the self-selected population.

This paper introduces a universal approach to seamlessly combine out-of-distribution (OOD) detection scores. These scores encompass a wide range of techniques that leverage the self-confidence of deep learning models and the anomalous behavior of features in the latent space. Not surprisingly, combining such a varied population using simple statistics proves inadequate. To overcome this challenge, we propose a quantile normalization to map these scores into p-values, effectively framing the problem into a multi-variate hypothesis test. Then, we combine these tests using established meta-analysis tools, resulting in a more effective detector with consolidated decision boundaries. Furthermore, we create a probabilistic interpretable criterion by mapping the final statistics into a distribution with known parameters. Through empirical investigation, we explore different types of shifts, each exerting varying degrees of impact on data. Our results demonstrate that our approach significantly improves overall robustness and performance across diverse OOD detection scenarios. Notably, our framework is easily extensible for future developments in detection scores and stands as the first to combine decision boundaries in this context. The code and artifacts associated with this work are publicly available\footnote{\url{https://github.com/edadaltocg/detectors}}.

The rapid scaling of large language models (LLMs) has raised concerns about the transparency and fair use of the pretraining data used for training them. Detecting such content is challenging due to the scale of the data and limited exposure of each instance during training. We propose ReCaLL (Relative Conditional Log-Likelihood), a novel membership inference attack (MIA) to detect LLMs' pretraining data by leveraging their conditional language modeling capabilities. ReCaLL examines the relative change in conditional log-likelihoods when prefixing target data points with non-member context. Our empirical findings show that conditioning member data on non-member prefixes induces a larger decrease in log-likelihood compared to non-member data. We conduct comprehensive experiments and show that ReCaLL achieves state-of-the-art performance on the WikiMIA dataset, even with random and synthetic prefixes, and can be further improved using an ensemble approach. Moreover, we conduct an in-depth analysis of LLMs' behavior with different membership contexts, providing insights into how LLMs leverage membership information for effective inference at both the sequence and token level.

This study builds upon our previous work by introducing a refined Inductive Conformal Martingale (ICM) approach for addressing Concept Drift (CD). Specifically, we enhance our previously proposed CAUTIOUS betting function to incorporate multiple density estimators for improving detection ability. We also combine this betting function with two base estimators that have not been previously utilized within the ICM framework: the Interpolated Histogram and Nearest Neighbor Density Estimators. We assess these extensions using both a single ICM and an ensemble of ICMs. For the latter, we conduct a comprehensive experimental investigation into the influence of the ensemble size on prediction accuracy and the number of available predictions. Our experimental results on four benchmark datasets demonstrate that the proposed approach surpasses our previous methodology in terms of performance while matching or in many cases exceeding that of three contemporary state-of-the-art techniques.

Integration of speech into healthcare has intensified privacy concerns due to its potential as a non-invasive biomarker containing individual biometric information. In response, speaker anonymization aims to conceal personally identifiable information while retaining crucial linguistic content. However, the application of anonymization techniques to pathological speech, a critical area where privacy is especially vital, has not been extensively examined. This study investigates anonymization's impact on pathological speech across over 2,700 speakers from multiple German institutions, focusing on privacy, pathological utility, and demographic fairness. We explore both deep-learning-based and signal processing-based anonymization methods, and document substantial privacy improvements across disorders-evidenced by equal error rate increases up to 1933%, with minimal overall impact on utility. Specific disorders such as Dysarthria, Dysphonia, and Cleft Lip and Palate experienced minimal utility changes, while Dysglossia showed slight improvements. Our findings underscore that the impact of anonymization varies substantially across different disorders. This necessitates disorder-specific anonymization strategies to optimally balance privacy with diagnostic utility. Additionally, our fairness analysis revealed consistent anonymization effects across most of the demographics. This study demonstrates the effectiveness of anonymization in pathological speech for enhancing privacy, while also highlighting the importance of customized and disorder-specific approaches to account for inversion attacks.

The manual examination of X-ray images for fractures is a time-consuming process that is prone to human error. In this work, we introduce a robust yet simple training loop for the classification of fractures, which significantly outperforms existing methods. Our method achieves superior performance in less than ten epochs and utilizes the latest dataset to deliver the best-performing model for this task. We emphasize the importance of training deep learning models responsibly and efficiently, as well as the critical role of selecting high-quality datasets.

Graph contrastive learning (GCL) aims to contrast positive-negative counterparts to learn the node embeddings, whereas graph data augmentation methods are employed to generate these positive-negative samples. The variation, quantity, and quality of negative samples compared to positive samples play crucial roles in learning meaningful embeddings for node classification downstream tasks. Less variation, excessive quantity, and low-quality negative samples cause the model to be overfitted for particular nodes, resulting in less robust models. To solve the overfitting problem in the GCL paradigm, this study proposes a novel Cumulative Sample Selection (CSS) algorithm by comprehensively considering negative samples' quality, variations, and quantity. Initially, three negative sample pools are constructed: easy, medium, and hard negative samples, which contain 25%, 50%, and 25% of the total available negative samples, respectively. Then, 10% negative samples are selected from each of these three negative sample pools for training the model. After that, a decision agent module evaluates model training results and decides whether to explore more negative samples from three negative sample pools by increasing the ratio or keep exploiting the current sampling ratio. The proposed algorithm is integrated into a proposed graph contrastive learning framework named NegAmplify. NegAmplify is compared with the SOTA methods on nine graph node classification datasets, with seven achieving better node classification accuracy with up to 2.86% improvement.

Flares are a well-studied aspect of the Sun's magnetic activity. Detecting and classifying solar flares can inform the analysis of contamination caused by stellar flares in exoplanet transmission spectra. In this paper, we present a standardized procedure to classify solar flares with the aid of supervised machine learning. Using flare data from the RHESSI mission and solar spectra from the HARPS-N instrument, we trained several supervised machine learning models, and found that the best performing algorithm is a C-Support Vector Machine (SVC) with non-linear kernels, specifically Radial Basis Functions (RBF). The best-trained model, SVC with RBF kernels, achieves an average aggregate accuracy score of 0.65, and categorical accuracy scores of over 0.70 for the no-flare and weak-flare classes, respectively. In comparison, a blind classification algorithm would have an accuracy score of 0.33. Testing showed that the model is able to detect and classify solar flares in entirely new data with different characteristics and distributions from those of the training set. Future efforts could focus on enhancing classification accuracy, investigating the efficacy of alternative models, particularly deep learning models, and incorporating more datasets to extend the application of this framework to stars that host exoplanets.