The development of larger models for medical image analysis has led to increased performance. However, it also affected our ability to explain and validate model decisions. Models can use non-relevant parts of images, also called spurious correlations or shortcuts, to obtain high performance on benchmark datasets but fail in real-world scenarios. In this work, we challenge the capacity of convolutional neural networks (CNN) to classify chest X-rays and eye fundus images while masking out clinically relevant parts of the image. We show that all models trained on the PadChest dataset, irrespective of the masking strategy, are able to obtain an Area Under the Curve (AUC) above random. Moreover, the models trained on full images obtain good performance on images without the region of interest (ROI), even superior to the one obtained on images only containing the ROI. We also reveal a possible spurious correlation in the Chaksu dataset while the performances are more aligned with the expectation of an unbiased model. We go beyond the performance analysis with the usage of the explainability method SHAP and the analysis of embeddings. We asked a radiology resident to interpret chest X-rays under different masking to complement our findings with clinical knowledge. Our code is available at https://github.com/TheoSourget/MMC_Masking and https://github.com/TheoSourget/MMC_Masking_EyeFundus
This study addresses the intricate challenge of decentralized output-feedback control for stochastic non-triangular nonlinear interconnected systems with unknown time-varying sensor sensitivity in a dynamic event-triggered context. The presence of stochastic disturbances, non-triangular structural uncertainties, and evolving sensor sensitivity distinguishes this problem of global asymptotic stability from conventional event-triggered control scenarios. Existing event-triggered control approaches with static event conditions encounter difficulties in simultaneously ensuring zero tracking/stabilization error and preventing the occurrence of Zeno behavior. In this work, we develop a novel solution to address this complex issue. Firstly, we establish a linear relationship between the state vector of each interconnected subsystem and two error vectors through a unique coordinate transformation. This transformation effectively handles the complexities introduced by non-triangular structural uncertainties. Secondly, we introduce a decentralized dynamic event-triggered output-feedback control strategy, which involves a state observer and a decentralized output-feedback controller. Unlike conventional event-triggered control methods with static event conditions, this strategy formulates a modified clock-based dynamic triggering mechanism by introducing an auxiliary variable that evolves based on predicted plant state values, while utilizing a clock variable to guarantee the existence of a positive lower bound on inter-execution times. Rigorous Lyapunov analysis confirms the global asymptotic stability in probability of the closed-loop system, with the states and the output of each local subsystem converging to the equilibrium at the origin in probability. Additionally, the existence of a minimal dwell-time between triggering instants is guaranteed.
Hepatocellular carcinoma (HCC) is a common type of liver cancer whose early-stage diagnosis is a common challenge, mainly due to the manual assessment of hematoxylin and eosin-stained whole slide images, which is a time-consuming process and may lead to variability in decision-making. For accurate detection of HCC, we propose a hybrid deep learning-based architecture that uses transfer learning to extract the features from pre-trained convolutional neural network (CNN) models and a classifier made up of a sequence of fully connected layers. This study uses a publicly available The Cancer Genome Atlas Hepatocellular Carcinoma (TCGA-LIHC)database (n=491) for model development and database of Kasturba Gandhi Medical College (KMC), India for validation. The pre-processing step involves patch extraction, colour normalization, and augmentation that results in 3920 patches for the TCGA dataset. The developed hybrid deep neural network consisting of a CNN-based pre-trained feature extractor and a customized artificial neural network-based classifier is trained using five-fold cross-validation. For this study, eight different state-of-the-art models are trained and tested as feature extractors for the proposed hybrid model. The proposed hybrid model with ResNet50-based feature extractor provided the sensitivity, specificity, F1-score, accuracy, and AUC of 100.00%, 100.00%, 100.00%, 100.00%, and 1.00, respectively on the TCGA database. On the KMC database, EfficientNetb3 resulted in the optimal choice of the feature extractor giving sensitivity, specificity, F1-score, accuracy, and AUC of 96.97, 98.85, 96.71, 96.71, and 0.99, respectively. The proposed hybrid models showed improvement in accuracy of 2% and 4% over the pre-trained models in TCGA-LIHC and KMC databases.
Psychophysical discrimination of structured light (SL) stimuli may be useful in screening for various macular disorders, including degenerative macular diseases. The circularly-oriented macular pigment optical density (coMPOD), calculated from the discrimination performance of SL-induced entoptic phenomena, may reveal a novel functional biomarker of macular health. In this study, we investigated the potential influence of eye dominance and testing order effects on SL-based stimulus perception, factors that potentially influence the sensitivity of screening tests based on SL technology. A total of 28 participants (aged 18-38 years) were selected for the study after undergoing a comprehensive eye examination. A psychophysical task was performed where various SL-based entoptic images with multiple azimuthal fringes rotating with a specific temporal frequency were projected onto the participants' retinas. By occluding the central areas of entoptic images, we measured the retinal eccentricity ($R$) of the perceivable area of the stimuli. The slope of the coMPOD profile ($a$-value) was calculated for each participant using a spatiotemporal sensitivity model that takes into account the perceptual threshold measurements of structured light stimuli with varying spatial densities and temporal frequencies. The Pearson correlation coefficient between eye dominance and testing order effects was $r=0.8$ ($p<0.01$). The Bland-Altman plots for both factors indicated zero bias. The results indicate repeatable measurements for both eyes, implying minimal impact from eye dominance and testing order on SL-based stimulus perception. The results provide a foundation for future studies exploring the clinical utility of SL tools in eye health.
Dynamic metasurface antennas (DMAs), surfaces patterned with reconfigurable metamaterial elements (meta-atoms) that couple waves from waveguides or cavities to free space, are a promising technology to realize 6G wireless base stations and access points with low cost and power consumption. Mutual coupling between the DMA's meta-atoms results in a non-linear dependence of the radiation pattern on the DMA configuration, significantly complicating modeling and optimization. Therefore, mutual coupling has to date been considered a vexing nuance that is frequently neglected in theoretical studies and deliberately mitigated in experimental prototypes. Here, we demonstrate the overlooked property of mutual coupling to boost the control over the DMA's radiation pattern. Based on a physics-compliant DMA model, we demonstrate that the radiation pattern's sensitivity to the DMA configuration significantly depends on the mutual coupling strength. We further evidence how the enhanced sensitivity under strong mutual coupling translates into a higher fidelity in radiation pattern synthesis, benefiting applications ranging from dynamic beamforming to end-to-end optimized sensing and imaging. Our insights suggest that DMA design should be fundamentally rethought to embrace the benefits of mutual coupling. We also discuss ensuing future research directions related to the frugal characterization of DMAs based on compact physics-compliant models.
The increasing reliance on deep computer vision models that process sensitive data has raised significant privacy concerns, particularly regarding the exposure of intermediate results in hidden layers. While traditional privacy risk assessment techniques focus on protecting overall model outputs, they often overlook vulnerabilities within these intermediate representations. Current privacy risk assessment techniques typically rely on specific attack simulations to assess risk, which can be computationally expensive and incomplete. This paper introduces a novel approach to measuring privacy risks in deep computer vision models based on the Degrees of Freedom (DoF) and sensitivity of intermediate outputs, without requiring adversarial attack simulations. We propose a framework that leverages DoF to evaluate the amount of information retained in each layer and combines this with the rank of the Jacobian matrix to assess sensitivity to input variations. This dual analysis enables systematic measurement of privacy risks at various model layers. Our experimental validation on real-world datasets demonstrates the effectiveness of this approach in providing deeper insights into privacy risks associated with intermediate representations.
The rapid advancement of generative models has introduced serious risks, including deepfake techniques for facial synthesis and editing. Traditional approaches rely on training classifiers and enhancing generalizability through various feature extraction techniques. Meanwhile, training-free detection methods address issues like limited data and overfitting by directly leveraging statistical properties from vision foundation models to distinguish between real and fake images. The current leading training-free approach, RIGID, utilizes DINOv2 sensitivity to perturbations in image space for detecting fake images, with fake image embeddings exhibiting greater sensitivity than those of real images. This observation prompts us to investigate how detection performance varies across model backbones, perturbation types, and datasets. Our experiments reveal that detection performance is closely linked to model robustness, with self-supervised (SSL) models providing more reliable representations. While Gaussian noise effectively detects general objects, it performs worse on facial images, whereas Gaussian blur is more effective due to potential frequency artifacts. To further improve detection, we introduce Contrastive Blur, which enhances performance on facial images, and MINDER (MINimum distance DetEctoR), which addresses noise type bias, balancing performance across domains. Beyond performance gains, our work offers valuable insights for both the generative and detection communities, contributing to a deeper understanding of model robustness property utilized for deepfake detection.
Expected Improvement (EI) is arguably the most widely used acquisition function in Bayesian optimization. However, it is often challenging to enhance the performance with EI due to its sensitivity to numerical precision. Previously, Hutter et al. (2009) tackled this problem by using Gaussian process trained on the log-transformed objective function and it was reported that this trick improves the predictive accuracy of GP, leading to substantially better performance. Although Hutter et al. (2009) offered the closed form of their EI, its intermediate derivation has not been provided so far. In this paper, we give a friendly derivation of their proposition.
Electricity grid's resiliency and climate change strongly impact one another due to an array of technical and policy-related decisions that impact both. This paper introduces a physics-informed machine learning-based framework to enhance grid's resiliency. Specifically, when encountering disruptive events, this paper designs remedial control actions to prevent blackouts. The proposed Physics-Guided Reinforcement Learning (PG-RL) framework determines effective real-time remedial line-switching actions, considering their impact on power balance, system security, and grid reliability. To identify an effective blackout mitigation policy, PG-RL leverages power-flow sensitivity factors to guide the RL exploration during agent training. Comprehensive evaluations using the Grid2Op platform demonstrate that incorporating physical signals into RL significantly improves resource utilization within electric grids and achieves better blackout mitigation policies - both of which are critical in addressing climate change.
Recent research efforts focus on reducing the computational and memory overheads of Large Language Models (LLMs) to make them feasible on resource-constrained devices. Despite advancements in compression techniques, non-linear operators like Softmax and Layernorm remain bottlenecks due to their sensitivity to quantization. We propose SoftmAP, a software-hardware co-design methodology that implements an integer-only low-precision Softmax using In-Memory Compute (IMC) hardware. Our method achieves up to three orders of magnitude improvement in the energy-delay product compared to A100 and RTX3090 GPUs, making LLMs more deployable without compromising performance.
Point cloud processing (PCP) encompasses tasks like reconstruction, denoising, registration, and segmentation, each often requiring specialized models to address unique task characteristics. While in-context learning (ICL) has shown promise across tasks by using a single model with task-specific demonstration prompts, its application to PCP reveals significant limitations. We identify inter-task and intra-task sensitivity issues in current ICL methods for PCP, which we attribute to inflexible sampling strategies lacking context adaptation at the point and prompt levels. To address these challenges, we propose MICAS, an advanced ICL framework featuring a multi-grained adaptive sampling mechanism tailored for PCP. MICAS introduces two core components: task-adaptive point sampling, which leverages inter-task cues for point-level sampling, and query-specific prompt sampling, which selects optimal prompts per query to mitigate intra-task sensitivity. To our knowledge, this is the first approach to introduce adaptive sampling tailored to the unique requirements of point clouds within an ICL framework. Extensive experiments show that MICAS not only efficiently handles various PCP tasks but also significantly outperforms existing methods. Notably, it achieves a remarkable $4.1\%$ improvement in the part segmentation task and delivers consistent gains across various PCP applications.
Stockart et al. (2025) recommend guidelines for best practices in the field of unconscious cognition. However, they condone the repeatedly criticized technique of excluding trials with high visibility ratings or of participants with high sensitivity for the critical stimulus. Based on standard signal detection theory for discrimination judgments, we show that post-hoc trial selection only isolates points of neutral response bias but remains consistent with uncomfortably high levels of sensitivity. We argue that post-hoc selection constitutes a sampling fallacy that capitalizes on chance, generates regression artifacts, and wrongly ascribes unconscious processing to stimulus conditions that may be far from indiscriminable. As an alternative, we advocate the study of functional dissociations, where direct (D) and indirect (I) measures are conceptualized as spanning up a two-dimensional D-I-space and where single, sensitivity, and double dissociations appear as distinct curve patterns. While Stockart et al.'s recommendations cover only a single line of that space where D is close to zero, functional dissociations can utilize the entire space, circumventing requirements like null visibility and exhaustive reliability, and allowing for the planful measurement of theoretically meaningful functional relationships between experimentally controlled variables.
Vectorized high-definition (HD) map is essential for autonomous driving, providing detailed and precise environmental information for advanced perception and planning. However, current map vectorization methods often exhibit deviations, and the existing evaluation metric for map vectorization lacks sufficient sensitivity to detect these deviations. To address these limitations, we propose integrating the philosophy of rasterization into map vectorization. Specifically, we introduce a new rasterization-based evaluation metric, which has superior sensitivity and is better suited to real-world autonomous driving scenarios. Furthermore, we propose MapVR (Map Vectorization via Rasterization), a novel framework that applies differentiable rasterization to vectorized outputs and then performs precise and geometry-aware supervision on rasterized HD maps. Notably, MapVR designs tailored rasterization strategies for various geometric shapes, enabling effective adaptation to a wide range of map elements. Experiments show that incorporating rasterization into map vectorization greatly enhances performance with no extra computational cost during inference, leading to more accurate map perception and ultimately promoting safer autonomous driving.
Effective monitoring of wildlife is critical for assessing biodiversity and ecosystem health, as declines in key species often signal significant environmental changes. Birds, particularly ground-nesting species, serve as important ecological indicators due to their sensitivity to environmental pressures. Camera traps have become indispensable tools for monitoring nesting bird populations, enabling data collection across diverse habitats. However, the manual processing and analysis of such data are resource-intensive, often delaying the delivery of actionable conservation insights. This study presents an AI-driven approach for real-time species detection, focusing on the curlew (Numenius arquata), a ground-nesting bird experiencing significant population declines. A custom-trained YOLOv10 model was developed to detect and classify curlews and their chicks using 3/4G-enabled cameras linked to the Conservation AI platform. The system processes camera trap data in real-time, significantly enhancing monitoring efficiency. Across 11 nesting sites in Wales, the model achieved high performance, with a sensitivity of 90.56%, specificity of 100%, and F1-score of 95.05% for curlew detections, and a sensitivity of 92.35%, specificity of 100%, and F1-score of 96.03% for curlew chick detections. These results demonstrate the capability of AI-driven monitoring systems to deliver accurate, timely data for biodiversity assessments, facilitating early conservation interventions and advancing the use of technology in ecological research.
Electron cryomicroscopy (cryo-EM) is a technique in structural biology used to reconstruct accurate volumetric maps of molecules. One step of the cryo-EM pipeline involves solving an inverse-problem. This inverse-problem, referred to as \textit{ab-initio} single-particle reconstruction, takes as input a collection of 2d-images -- each a projection of a molecule from an unknown viewing-angle -- and attempts to reconstruct the 3d-volume representing the underlying molecular density. Most methods for solving this inverse-problem search for a solution which optimizes a posterior likelihood of generating the observed image-data, given the reconstructed volume. Within this framework, it is natural to study the Hessian of the log-likelihood: the eigenvectors and eigenvalues of the Hessian determine how the likelihood changes with respect to perturbations in the solution, and can give insight into the sensitivity of the solution to aspects of the input. In this paper we describe a simple strategy for estimating the smallest eigenvalues and eigenvectors (i.e., the `softest modes') of the Hessian of the log-likelihood for the \textit{ab-initio} single-particle reconstruction problem. This strategy involves rewriting the log-likelihood as a 3d-integral. This interpretation holds in the low-noise limit, as well as in many practical scenarios which allow for noise-marginalization. Once we have estimated the softest modes, we can use them to perform many kinds of sensitivity analysis. For example, we can determine which parts of the reconstructed volume are trustworthy, and which are unreliable, and how this unreliability might depend on the data-set and the imaging parameters. We believe that this kind of analysis can be used alongside more traditional strategies for sensitivity analysis, as well as in other applications, such as free-energy estimation.
Driven by the critical challenges in biomanufacturing, including high complexity and high uncertainty, we propose a comprehensive and computationally efficient sensitivity analysis framework for general nonlinear policy-augmented knowledge graphical (pKG) hybrid models that characterize the risk- and science-based understandings of underlying stochastic decision process mechanisms. The criticality of each input (i.e., random factors, policy parameters, and model parameters) is measured by applying Shapley value (SV) sensitivity analysis to pKG (called SV-pKG), accounting for process causal interdependences. To quickly assess the SV for heavily instrumented bioprocesses, we approximate their dynamics with linear Gaussian pKG models and improve the SV estimation efficiency by utilizing the linear Gaussian properties. In addition, we propose an effective permutation sampling method with TFWW transformation and variance reduction techniques, namely the quasi-Monte Carlo and antithetic sampling methods, to further improve the sampling efficiency and estimation accuracy of SV for both general nonlinear and linear Gaussian pKG models. Our proposed framework can benefit efficient interpretation and support stable optimal process control in biomanufacturing.
Autism and Attention-Deficit Hyperactivity Disorder (ADHD) are two of the most commonly observed neurodevelopmental conditions in childhood. Providing a specific computational assessment to distinguish between the two can prove difficult and time intensive. Given the high prevalence of their co-occurrence, there is a need for scalable and accessible methods for distinguishing the co-occurrence of autism and ADHD from individual diagnoses. The first step is to identify a core set of features that can serve as the basis for behavioral feature extraction. We trained machine learning models on data from the National Survey of Children's Health to identify behaviors to target as features in automated clinical decision support systems. A model trained on the binary task of distinguishing either developmental delay (autism or ADHD) vs. neither achieved sensitivity >92% and specificity >94%, while a model trained on the 4-way classification task of autism vs. ADHD vs. both vs. none demonstrated >65% sensitivity and >66% specificity. While the performance of the binary model was respectable, the relatively low performance in the differential classification of autism and ADHD highlights the challenges that persist in achieving specificity within clinical decision support tools for developmental delays. Nevertheless, this study demonstrates the potential of applying behavioral questionnaires not traditionally used for clinical purposes towards supporting digital screening assessments for pediatric developmental delays.
Longitudinal analyses are increasingly used in clinical studies as they allow the study of subtle changes over time within the same subjects. In most of these studies, it is necessary to align all the images studied to a common reference by registering them to a template. In the study of white matter using the recently developed fixel-based analysis (FBA) method, this registration is important, in particular because the fiber bundle cross-section metric is a direct measure of this registration. In the vast majority of longitudinal FBA studies described in the literature, sessions acquired for a same subject are directly independently registered to the template. However, it has been shown in T1-based morphometry that a 2-step registration through an intra-subject average can be advantageous in longitudinal analyses. In this work, we propose an implementation of this 2-step registration method in a typical longitudinal FBA aimed at investigating the evolution of white matter changes in Alzheimer's disease (AD). We compared at the fixel level the mean absolute effect and standard deviation yielded by this registration method and by a direct registration, as well as the results obtained with each registration method for the study of AD in both fixelwise and tract-based analyses. We found that the 2-step method reduced the variability of the measurements and thus enhanced statistical power in both types of analyses.
Assessment of the sympathetic nervous system (SNS) is one of the major approaches for studying affective states. Skin nerve activity (SKNA) derived from high-frequency components of electrocardiogram (ECG) signals has been a promising surrogate for assessing the SNS. However, current SKNA analysis tools have shown high variability across study protocols and experiments. Hence, we propose a time-varying spectral approach based on SKNA to assess the SNS with higher sensitivity and reliability. We collected ECG signals at a sampling frequency of 10 KHz from sixteen subjects who underwent various SNS stimulations. Our spectral analysis revealed that frequency bands between 150 - 1,000 Hz showed significant increases in power during SNS stimulations. Using this information, we developed a time-varying index of sympathetic function measurement based on SKNA, termed, Time-Varying Skin Nerve Activity (TVSKNA). TVSKNA is calculated in three steps: time-frequency decomposition, reconstruction using selected frequency bands, and smoothing. TVSKNA indices exhibited generally higher Youden's J, balanced accuracy, and area under the receiver operating characteristic curve, indicating higher sensitivity. The coefficient of variance was lower with TVSKNA indices for most SNS tasks. TVSKNA can serve as a highly sensitive and reliable marker of quantitative assessment of sympathetic function, especially during emotion and stress.
Large language models (LLMs) are known to exhibit demographic biases, yet few studies systematically evaluate these biases across multiple datasets or account for confounding factors. In this work, we examine LLM alignment with human annotations in five offensive language datasets, comprising approximately 220K annotations. Our findings reveal that while demographic traits, particularly race, influence alignment, these effects are inconsistent across datasets and often entangled with other factors. Confounders -- such as document difficulty, annotator sensitivity, and within-group agreement -- account for more variation in alignment patterns than demographic traits alone. Specifically, alignment increases with higher annotator sensitivity and group agreement, while greater document difficulty corresponds to reduced alignment. Our results underscore the importance of multi-dataset analyses and confounder-aware methodologies in developing robust measures of demographic bias in LLMs.