With large language models (LLMs) poised to become embedded in our daily lives, questions are starting to be raised about the dataset(s) they learned from. These questions range from potential bias or misinformation LLMs could retain from their training data to questions of copyright and fair use of human-generated text. However, while these questions emerge, developers of the recent state-of-the-art LLMs become increasingly reluctant to disclose details on their training corpus. We here introduce the task of document-level membership inference for real-world LLMs, i.e. inferring whether the LLM has seen a given document during training or not. First, we propose a procedure for the development and evaluation of document-level membership inference for LLMs by leveraging commonly used data sources for training and the model release date. We then propose a practical, black-box method to predict document-level membership and instantiate it on OpenLLaMA-7B with both books and academic papers. We show our methodology to perform very well, reaching an impressive AUC of 0.856 for books and 0.678 for papers. We then show our approach to outperform the sentence-level membership inference attacks used in the privacy literature for the document-level membership task. We finally evaluate whether smaller models might be less sensitive to document-level inference and show OpenLLaMA-3B to be approximately as sensitive as OpenLLaMA-7B to our approach. Taken together, our results show that accurate document-level membership can be inferred for LLMs, increasing the transparency of technology poised to change our lives.

When deploying machine learning systems to the wild, it is highly desirable for them to effectively leverage prior knowledge to the unfamiliar domain while also firing alarms to anomalous inputs. In order to address these requirements, Universal Domain Adaptation (UniDA) has emerged as a novel research area in computer vision, focusing on achieving both adaptation ability and robustness (i.e., the ability to detect out-of-distribution samples). While UniDA has led significant progress in computer vision, its application on language input still needs to be explored despite its feasibility. In this paper, we propose a comprehensive benchmark for natural language that offers thorough viewpoints of the model's generalizability and robustness. Our benchmark encompasses multiple datasets with varying difficulty levels and characteristics, including temporal shifts and diverse domains. On top of our testbed, we validate existing UniDA methods from computer vision and state-of-the-art domain adaptation techniques from NLP literature, yielding valuable findings: We observe that UniDA methods originally designed for image input can be effectively transferred to the natural language domain while also underscoring the effect of adaptation difficulty in determining the model's performance.

Large language models (LLMs) are transforming the ways the general public accesses and consumes information. Their influence is particularly pronounced in pivotal sectors like healthcare, where lay individuals are increasingly appropriating LLMs as conversational agents for everyday queries. While LLMs demonstrate impressive language understanding and generation proficiencies, concerns regarding their safety remain paramount in these high-stake domains. Moreover, the development of LLMs is disproportionately focused on English. It remains unclear how these LLMs perform in the context of non-English languages, a gap that is critical for ensuring equity in the real-world use of these systems.This paper provides a framework to investigate the effectiveness of LLMs as multi-lingual dialogue systems for healthcare queries. Our empirically-derived framework XlingEval focuses on three fundamental criteria for evaluating LLM responses to naturalistic human-authored health-related questions: correctness, consistency, and verifiability. Through extensive experiments on four major global languages, including English, Spanish, Chinese, and Hindi, spanning three expert-annotated large health Q&A datasets, and through an amalgamation of algorithmic and human-evaluation strategies, we found a pronounced disparity in LLM responses across these languages, indicating a need for enhanced cross-lingual capabilities. We further propose XlingHealth, a cross-lingual benchmark for examining the multilingual capabilities of LLMs in the healthcare context. Our findings underscore the pressing need to bolster the cross-lingual capacities of these models, and to provide an equitable information ecosystem accessible to all.

Prompt Tuning is emerging as a scalable and cost-effective method to fine-tune Pretrained Language Models (PLMs), which are often referred to as Large Language Models (LLMs). This study benchmarks the performance and computational efficiency of Prompt Tuning and baselines for multi-label text classification. This is applied to the challenging task of classifying companies into an investment firm's proprietary industry taxonomy, supporting their thematic investment strategy. Text-to-text classification is frequently reported to outperform task-specific classification heads, but has several limitations when applied to a multi-label classification problem where each label consists of multiple tokens: (a) Generated labels may not match any label in the label taxonomy; (b) The fine-tuning process lacks permutation invariance and is sensitive to the order of the provided labels; (c) The model provides binary decisions rather than appropriate confidence scores. Limitation (a) is addressed by applying constrained decoding using Trie Search, which slightly improves classification performance. All limitations (a), (b), and (c) are addressed by replacing the PLM's language head with a classification head, which is referred to as Prompt Tuned Embedding Classification (PTEC). This improves performance significantly, while also reducing computational costs during inference. In our industrial application, the training data is skewed towards well-known companies. We confirm that the model's performance is consistent across both well-known and less-known companies. Our overall results indicate the continuing need to adapt state-of-the-art methods to domain-specific tasks, even in the era of PLMs with strong generalization abilities. We release our codebase and a benchmarking dataset at https://github.com/EQTPartners/PTEC.

As the complexity of machine learning (ML) models increases and their application in different (and critical) domains grows, there is a strong demand for more interpretable and trustworthy ML. A direct, model-agnostic, way to interpret such models is to train surrogate models, such as rule sets and decision trees, that sufficiently approximate the original ones while being simpler and easier-to-explain. Yet, rule sets can become very lengthy, with many if-else statements, and decision tree depth grows rapidly when accurately emulating complex ML models. In such cases, both approaches can fail to meet their core goal, providing users with model interpretability. To tackle this, we propose DeforestVis, a visual analytics tool that offers user-friendly summarization of the behavior of complex ML models by providing surrogate decision stumps (one-level decision trees) generated with the adaptive boosting (AdaBoost) technique. DeforestVis helps users to explore the complexity vs fidelity trade-off by incrementally generating more stumps, creating attribute-based explanations with weighted stumps to justify decision making, and analyzing the impact of rule overriding on training instance allocation between one or more stumps. An independent test set allows users to monitor the effectiveness of manual rule changes and form hypotheses based on case-by-case analyses. We show the applicability and usefulness of DeforestVis with two use cases and expert interviews with data analysts and model developers.

We study the key framework of learning with abstention in the multi-class classification setting. In this setting, the learner can choose to abstain from making a prediction with some pre-defined cost. We present a series of new theoretical and algorithmic results for this learning problem in the predictor-rejector framework. We introduce several new families of surrogate losses for which we prove strong non-asymptotic and hypothesis set-specific consistency guarantees, thereby resolving positively two existing open questions. These guarantees provide upper bounds on the estimation error of the abstention loss function in terms of that of the surrogate loss. We analyze both a single-stage setting where the predictor and rejector are learned simultaneously and a two-stage setting crucial in applications, where the predictor is learned in a first stage using a standard surrogate loss such as cross-entropy. These guarantees suggest new multi-class abstention algorithms based on minimizing these surrogate losses. We also report the results of extensive experiments comparing these algorithms to the current state-of-the-art algorithms on CIFAR-10, CIFAR-100 and SVHN datasets. Our results demonstrate empirically the benefit of our new surrogate losses and show the remarkable performance of our broadly applicable two-stage abstention algorithm.

We propose a novel statistical method for testing the results of anomaly detection (AD) under domain adaptation (DA), which we call CAD-DA -- controllable AD under DA. The distinct advantage of the CAD-DA lies in its ability to control the probability of misidentifying anomalies under a pre-specified level $\alpha$ (e.g., 0.05). The challenge within this DA setting is the necessity to account for the influence of DA to ensure the validity of the inference results. Our solution to this challenge leverages the concept of conditional Selective Inference to handle the impact of DA. To our knowledge, this is the first work capable of conducting a valid statistical inference within the context of DA. We evaluate the performance of the CAD-DA method on both synthetic and real-world datasets.

Vessel structure is one of the most important parts of the retina which physicians can detect many diseases by analysing its features. Localization of blood vessels in retina images is an important process in medical image analysis. This process is also more challenging with the presence of bright and dark lesions. In this thesis, two automated vessel localization methods to handle both healthy and unhealthy (pathological) retina images are analyzed. Each method consists of two major steps and the second step is the same in the two methods. In the first step, an algorithm is used to decrease the effect of bright lesions. In Method 1, this algorithm is based on K- Means segmentation, and in Method 2, it is based on a regularization procedure. In the second step of both methods, a multi-scale line operator is used to localize the line-shaped vascular structures and ignore the dark lesions which are generally assumed to have irregular patterns. After the introduction of the methods, a detailed quantitative and qualitative comparison of the methods with one another as well as the state-of-the-art solutions in the literature based on the segmentation results on the images of the two publicly available datasets, DRIVE and STARE, is reported. The results demonstrate that the methods are highly comparable with other solutions.

Graphs are mathematical tools that can be used to represent complex real-world interconnected systems, such as financial markets and social networks. Hence, machine learning (ML) over graphs has attracted significant attention recently. However, it has been demonstrated that ML over graphs amplifies the already existing bias towards certain under-represented groups in various decision-making problems due to the information aggregation over biased graph structures. Faced with this challenge, here we take a fresh look at the problem of bias mitigation in graph-based learning by borrowing insights from graph signal processing. Our idea is to introduce predesigned graph filters within an ML pipeline to reduce a novel unsupervised bias measure, namely the correlation between sensitive attributes and the underlying graph connectivity. We show that the optimal design of said filters can be cast as a convex problem in the graph spectral domain. We also formulate a linear programming (LP) problem informed by a theoretical bias analysis, which attains a closed-form solution and leads to a more efficient fairness-aware graph filter. Finally, for a design whose degrees of freedom are independent of the input graph size, we minimize the bias metric over the family of polynomial graph convolutional filters. Our optimal filter designs offer complementary strengths to explore favorable fairness-utility-complexity tradeoffs. For performance evaluation, we conduct extensive and reproducible node classification experiments over real-world networks. Our results show that the proposed framework leads to better fairness measures together with similar utility compared to state-of-the-art fairness-aware baselines.

Object recognition is a crucial step in perception systems for autonomous and intelligent vehicles, as evidenced by the numerous research works in the topic. In this paper, object recognition is explored by using multisensory and multimodality approaches, with the intention of reducing the false positive rate (FPR). The reduction of the FPR becomes increasingly important in perception systems since the misclassification of an object can potentially cause accidents. In particular, this work presents a strategy through Bayesian inference to reduce the FPR considering the likelihood function as a cumulative distribution function from Gaussian kernel density estimations, and the prior probabilities as cumulative functions of normalized histograms. The validation of the proposed methodology is performed on the KITTI dataset using deep networks (DenseNet, NasNet, and EfficientNet), and recent 3D point cloud networks (PointNet, and PintNet++), by considering three object-categories (cars, cyclists, pedestrians) and the RGB and LiDAR sensor modalities.