Despite their impressive capabilities, large language models (LLMs) can be prone to generating hallucinations -- undesirable outputs that are incorrect, unfaithful, or inconsistent with respect to the inputs (or the output itself) [1]. These unreliable behaviors pose significant risks for adopting LLMs in real-world applications. Challenges in detecting hallucinations lie, among other things, in hallucinations taking different forms, being context-dependent and sometimes being in conflict with other desirable properties of generated text [2, 3]. Hallucinations may be harmless in some contexts, but can be undesired or potentially dangerous in other applications (e.g., erroneous medical advice). Detecting and quantifying hallucination risk is thus a critical capability to enable safe applications of LLMs and improve generated outputs. Prior work has proposed various approaches for detecting and mitigating hallucinations in LLM-generated outputs, including verifying faithfulness to inputs [4], assessing internal coherence [5], consulting external knowledge sources [6], and quantifying model uncertainty [2, 3, 7, 8]. However, deploying these methods in production settings is far from trivial due to several challenges: First, there is limited comparative evaluation illuminating how different detection methods perform. Second, existing approaches for detecting hallucinations differ greatly in their computational demands, and guidelines are lacking on cost-effectiveness trade-offs to inform method selection for real-world applications with constraints. Third, hallucination detection in the real world often requires careful consideration of risks and false positive/negative trade-offs, requiring methods to provide well-calibrated probability scores.

The ability to detect out-of-distribution (OOD) inputs is critical to guarantee the reliability of classification models deployed in an open environment. A fundamental challenge in OOD detection is that a discriminative classifier is typically trained to estimate the posterior probability p(y|z) for class y given an input z, but lacks the explicit likelihood estimation of p(z) ideally needed for OOD detection. While numerous OOD scoring functions have been proposed for classification models, these estimate scores are often heuristic-driven and cannot be rigorously interpreted as likelihood. To bridge the gap, we propose Intrinsic Likelihood (INK), which offers rigorous likelihood interpretation to modern discriminative-based classifiers. Specifically, our proposed INK score operates on the constrained latent embeddings of a discriminative classifier, which are modeled as a mixture of hyperspherical embeddings with constant norm. We draw a novel connection between the hyperspherical distribution and the intrinsic likelihood, which can be effectively optimized in modern neural networks. Extensive experiments on the OpenOOD benchmark empirically demonstrate that INK establishes a new state-of-the-art in a variety of OOD detection setups, including both far-OOD and near-OOD. Code is available at https://github.com/deeplearning-wisc/ink.

Machine learning models can leak private information about their training data, but the standard methods to measure this risk, based on membership inference attacks (MIAs), have a major limitation. They only check if a given data point \textit{exactly} matches a training point, neglecting the potential of similar or partially overlapping data revealing the same private information. To address this issue, we introduce the class of range membership inference attacks (RaMIAs), testing if the model was trained on any data in a specified range (defined based on the semantics of privacy). We formulate the RaMIAs game and design a principled statistical test for its complex hypotheses. We show that RaMIAs can capture privacy loss more accurately and comprehensively than MIAs on various types of data, such as tabular, image, and language. RaMIA paves the way for a more comprehensive and meaningful privacy auditing of machine learning algorithms.

Various machine learning (ML)-based in-situ monitoring systems have been developed to detect laser additive manufacturing (LAM) process anomalies and defects. Multimodal fusion can improve in-situ monitoring performance by acquiring and integrating data from multiple modalities, including visual and audio data. However, multimodal fusion employs multiple sensors of different types, which leads to higher hardware, computational, and operational costs. This paper proposes a cross-modality knowledge transfer (CMKT) methodology that transfers knowledge from a source to a target modality for LAM in-situ monitoring. CMKT enhances the usefulness of the features extracted from the target modality during the training phase and removes the sensors of the source modality during the prediction phase. This paper proposes three CMKT methods: semantic alignment, fully supervised mapping, and semi-supervised mapping. Semantic alignment establishes a shared encoded space between modalities to facilitate knowledge transfer. It utilizes a semantic alignment loss to align the distributions of the same classes (e.g., visual defective and audio defective classes) and a separation loss to separate the distributions of different classes (e.g., visual defective and audio defect-free classes). The two mapping methods transfer knowledge by deriving the features of one modality from the other modality using fully supervised and semi-supervised learning. The proposed CMKT methods were implemented and compared with multimodal audio-visual fusion in an LAM in-situ anomaly detection case study. The semantic alignment method achieves a 98.4% accuracy while removing the audio modality during the prediction phase, which is comparable to the accuracy of multimodal fusion (98.2%).

Atrial fibrillation (AF) is characterized by irregular electrical impulses originating in the atria, which can lead to severe complications and even death. Due to the intermittent nature of the AF, early and timely monitoring of AF is critical for patients to prevent further exacerbation of the condition. Although ambulatory ECG Holter monitors provide accurate monitoring, the high cost of these devices hinders their wider adoption. Current mobile-based AF detection systems offer a portable solution. However, these systems have various applicability issues, such as being easily affected by environmental factors and requiring significant user effort. To overcome the above limitations, we present AcousAF, a novel AF detection system based on acoustic sensors of smartphones. Particularly, we explore the potential of pulse wave acquisition from the wrist using smartphone speakers and microphones. In addition, we propose a well-designed framework comprised of pulse wave probing, pulse wave extraction, and AF detection to ensure accurate and reliable AF detection. We collect data from 20 participants utilizing our custom data collection application on the smartphone. Extensive experimental results demonstrate the high performance of our system, with 92.8% accuracy, 86.9% precision, 87.4% recall, and 87.1% F1 Score.

In this paper the accuracy and robustness of quality measures for the assessment of machine learning models are investigated. The prediction quality of a machine learning model is evaluated model-independent based on a cross-validation approach, where the approximation error is estimated for unknown data. The presented measures quantify the amount of explained variation in the model prediction. The reliability of these measures is assessed by means of several numerical examples, where an additional data set for the verification of the estimated prediction error is available. Furthermore, the confidence bounds of the presented quality measures are estimated and local quality measures are derived from the prediction residuals obtained by the cross-validation approach.

The emergence of federated learning (FL) presents a promising approach to leverage decentralized data while preserving privacy. Furthermore, the combination of FL and anomaly detection is particularly compelling because it allows for detecting rare and critical anomalies (usually also rare in locally gathered data) in sensitive data from multiple sources, such as cybersecurity and healthcare. However, benchmarking the performance of anomaly detection methods in FL environments remains an underexplored area. This paper introduces FedAD-Bench, a unified benchmark for evaluating unsupervised anomaly detection algorithms within the context of FL. We systematically analyze and compare the performance of recent deep learning anomaly detection models under federated settings, which were typically assessed solely in centralized settings. FedAD-Bench encompasses diverse datasets and metrics to provide a holistic evaluation. Through extensive experiments, we identify key challenges such as model aggregation inefficiencies and metric unreliability. We present insights into FL's regularization effects, revealing scenarios in which it outperforms centralized approaches due to its inherent ability to mitigate overfitting. Our work aims to establish a standardized benchmark to guide future research and development in federated anomaly detection, promoting reproducibility and fair comparison across studies.

We propose using natural language outlines as a novel modality and interaction surface for providing AI assistance to developers throughout the software development process. An NL outline for a code function comprises multiple statements written in concise prose, which partition the code and summarize its main ideas in the style of literate programming. Crucially, we find that modern LLMs can generate accurate and high-quality NL outlines in practice. Moreover, NL outlines enable a bidirectional sync between code and NL, allowing changes in one to be automatically reflected in the other. We discuss many use cases for NL outlines: they can accelerate understanding and navigation of code and diffs, simplify code maintenance, augment code search, steer code generation, and more. We then propose and compare multiple LLM prompting techniques for generating outlines and ask professional developers to judge outline quality. Finally, we present two case studies applying NL outlines toward code review and the difficult task of malware detection.

Algorithmic bias in healthcare mirrors existing data biases. However, the factors driving unfairness are not always known. Medical devices capture significant amounts of data but are prone to errors; for instance, pulse oximeters overestimate the arterial oxygen saturation of darker-skinned individuals, leading to worse outcomes. The impact of this bias in machine learning (ML) models remains unclear. This study addresses the technical challenges of quantifying the impact of medical device bias in downstream ML. Our experiments compare a "perfect world", without pulse oximetry bias, using SaO2 (blood-gas), to the "actual world", with biased measurements, using SpO2 (pulse oximetry). Under this counterfactual design, two models are trained with identical data, features, and settings, except for the method of measuring oxygen saturation: models using SaO2 are a "control" and models using SpO2 a "treatment". The blood-gas oximetry linked dataset was a suitable test-bed, containing 163,396 nearly-simultaneous SpO2 - SaO2 paired measurements, aligned with a wide array of clinical features and outcomes. We studied three classification tasks: in-hospital mortality, respiratory SOFA score in the next 24 hours, and SOFA score increase by two points. Models using SaO2 instead of SpO2 generally showed better performance. Patients with overestimation of O2 by pulse oximetry of > 3% had significant decreases in mortality prediction recall, from 0.63 to 0.59, P < 0.001. This mirrors clinical processes where biased pulse oximetry readings provide clinicians with false reassurance of patients' oxygen levels. A similar degradation happened in ML models, with pulse oximetry biases leading to more false negatives in predicting adverse outcomes.

Path loss prediction for wireless communications is highly dependent on the local environment. Propagation models including clutter information have been shown to significantly increase model accuracy. This paper explores the application of deep learning to satellite imagery to identify environmental clutter types automatically. Recognizing these clutter types has numerous uses, but our main application is to use clutter information to enhance propagation prediction models. Knowing the type of obstruction (tree, building, and further classifications) can improve the prediction accuracy of key propagation metrics such as path loss.