Deep prior-based approaches have demonstrated remarkable success in blind motion deblurring (BMD) recently. These methods, however, are often limited by the high non-convexity of the underlying optimization process in BMD, which leads to extreme sensitivity to the initial blur kernel. To address this issue, we propose a novel framework for BMD that leverages a deep generative model to encode the kernel prior and induce a better initialization for the blur kernel. Specifically, we pre-train a kernel generator based on a generative adversarial network (GAN) to aptly characterize the kernel's prior distribution, as well as a kernel initializer to provide a well-informed and high-quality starting point for kernel estimation. By combining these two components, we constrain the BMD solution within a compact latent kernel manifold, thus alleviating the aforementioned sensitivity for kernel initialization. Notably, the kernel generator and initializer are designed to be easily integrated with existing BMD methods in a plug-and-play manner, enhancing their overall performance. Furthermore, we extend our approach to tackle blind non-uniform motion deblurring without the need for additional priors, achieving state-of-the-art performance on challenging benchmark datasets. The source code is available at https://github.com/dch0319/GLKM-Deblur.
Large Language Models (LLMs) have shown remarkable performance across diverse tasks without domain-specific training, fueling interest in their potential for time-series forecasting. While LLMs have shown potential in zero-shot forecasting through prompting alone, recent studies suggest that LLMs lack inherent effectiveness in forecasting. Given these conflicting findings, a rigorous validation is essential for drawing reliable conclusions. In this paper, we evaluate the effectiveness of LLMs as zero-shot forecasters compared to state-of-the-art domain-specific models. Our experiments show that LLM-based zero-shot forecasters often struggle to achieve high accuracy due to their sensitivity to noise, underperforming even simple domain-specific models. We have explored solutions to reduce LLMs' sensitivity to noise in the zero-shot setting, but improving their robustness remains a significant challenge. Our findings suggest that rather than emphasizing zero-shot forecasting, a more promising direction would be to focus on fine-tuning LLMs to better process numerical sequences. Our experimental code is available at https://github.com/junwoopark92/revisiting-LLMs-zeroshot-forecaster.
Computational inverse problems for biomedical simulators suffer from limited data and relatively high parameter dimensionality. This often requires sensitivity analysis, where parameters of the model are ranked based on their influence on the specific quantities of interest. This is especially important for simulators used to build medical digital twins, as the amount of data is typically limited. For expensive models, such as blood flow models, emulation is employed to expedite the simulation time. Parameter ranking and fixing using sensitivity analysis are often heuristic, though, and vary with the specific application or simulator used. The present study provides an innovative solution to this problem by leveraging polynomial chaos expansions (PCEs) for both multioutput global sensitivity analysis and formal parameter identifiability. For the former, we use dimension reduction to efficiently quantify time-series sensitivity of a one-dimensional pulmonary hemodynamics model. We consider both Windkessel and structured tree boundary conditions. We then use PCEs to construct profile-likelihood confidence intervals to formally assess parameter identifiability, and show how changes in experimental design improve identifiability. Our work presents a novel approach to determining parameter identifiability and leverages a common emulation strategy for enabling profile-likelihood analysis in problems governed by partial differential equations.
This paper studies the identification of the average treatment effect on the treated (ATT) under unconfoundedness when covariate overlap is partial. A formal diagnostic is proposed to characterize empirical support -- the subset of the covariate space where ATT is point-identified due to the presence of comparable untreated units. Where support is absent, standard estimators remain computable but cease to identify meaningful causal parameters. A general sensitivity framework is developed, indexing identified sets by curvature constraints on the selection mechanism. This yields a structural selection frontier tracing the trade-off between assumption strength and inferential precision. Two diagnostic statistics are introduced: the minimum assumption strength for sign identification (MAS-SI), and a fragility index that quantifies the minimal deviation from ignorability required to overturn qualitative conclusions. Applied to the LaLonde (1986) dataset, the framework reveals that nearly half the treated strata lack empirical support, rendering the ATT undefined in those regions. Simulations confirm that ATT estimates may be stable in magnitude yet fragile in epistemic content. These findings reframe overlap not as a regularity condition but as a prerequisite for identification, and recast sensitivity analysis as integral to empirical credibility rather than auxiliary robustness.
Learning with relational and network-structured data is increasingly vital in sensitive domains where protecting the privacy of individual entities is paramount. Differential Privacy (DP) offers a principled approach for quantifying privacy risks, with DP-SGD emerging as a standard mechanism for private model training. However, directly applying DP-SGD to relational learning is challenging due to two key factors: (i) entities often participate in multiple relations, resulting in high and difficult-to-control sensitivity; and (ii) relational learning typically involves multi-stage, potentially coupled (interdependent) sampling procedures that make standard privacy amplification analyses inapplicable. This work presents a principled framework for relational learning with formal entity-level DP guarantees. We provide a rigorous sensitivity analysis and introduce an adaptive gradient clipping scheme that modulates clipping thresholds based on entity occurrence frequency. We also extend the privacy amplification results to a tractable subclass of coupled sampling, where the dependence arises only through sample sizes. These contributions lead to a tailored DP-SGD variant for relational data with provable privacy guarantees. Experiments on fine-tuning text encoders over text-attributed network-structured relational data demonstrate the strong utility-privacy trade-offs of our approach. Our code is available at https://github.com/Graph-COM/Node_DP.
In recent years, researchers have started analyzing the cultural sensitivity of LLMs. In this respect, Etiquettes have been an active area of research. Etiquettes are region-specific and are an essential part of the culture of a region; hence, it is imperative to make LLMs sensitive to etiquettes. However, there needs to be more resources in evaluating LLMs for their understanding and bias with regard to etiquettes. In this resource paper, we introduce EtiCor++, a corpus of etiquettes worldwide. We introduce different tasks for evaluating LLMs for knowledge about etiquettes across various regions. Further, we introduce various metrics for measuring bias in LLMs. Extensive experimentation with LLMs shows inherent bias towards certain regions.
Large language models (LLMs) are increasingly becoming valuable to corporate data management due to their ability to process text from various document formats and facilitate user interactions through natural language queries. However, LLMs must consider the sensitivity of information when communicating with employees, especially given access restrictions. Simple filtering based on user clearance levels can pose both performance and privacy challenges. To address this, we propose the concept of sensitivity awareness (SA), which enables LLMs to adhere to predefined access rights rules. In addition, we developed a benchmarking environment called ACCESS DENIED INC to evaluate SA. Our experimental findings reveal significant variations in model behavior, particularly in managing unauthorized data requests while effectively addressing legitimate queries. This work establishes a foundation for benchmarking sensitivity-aware language models and provides insights to enhance privacy-centric AI systems in corporate environments.
Data attribution methods, which quantify the influence of individual training data points on a machine learning model, have gained increasing popularity in data-centric applications in modern AI. Despite a recent surge of new methods developed in this space, the impact of hyperparameter tuning in these methods remains under-explored. In this work, we present the first large-scale empirical study to understand the hyperparameter sensitivity of common data attribution methods. Our results show that most methods are indeed sensitive to certain key hyperparameters. However, unlike typical machine learning algorithms -- whose hyperparameters can be tuned using computationally-cheap validation metrics -- evaluating data attribution performance often requires retraining models on subsets of training data, making such metrics prohibitively costly for hyperparameter tuning. This poses a critical open challenge for the practical application of data attribution methods. To address this challenge, we advocate for better theoretical understandings of hyperparameter behavior to inform efficient tuning strategies. As a case study, we provide a theoretical analysis of the regularization term that is critical in many variants of influence function methods. Building on this analysis, we propose a lightweight procedure for selecting the regularization value without model retraining, and validate its effectiveness across a range of standard data attribution benchmarks. Overall, our study identifies a fundamental yet overlooked challenge in the practical application of data attribution, and highlights the importance of careful discussion on hyperparameter selection in future method development.
Code Sensitivity refers to the ability of Code LLMs to recognize and respond to details changes in problem descriptions. While current code benchmarks and instruction data focus on difficulty and diversity, sensitivity is overlooked. We first introduce the CTF-Code benchmark, constructed using counterfactual perturbations, minimizing input changes while maximizing output changes. The evaluation shows that many LLMs have a more than 10\% performance drop compared to the original problems. To fully utilize sensitivity, CTF-Instruct, an incremental instruction fine-tuning framework, extends on existing data and uses a selection mechanism to meet the three dimensions of difficulty, diversity, and sensitivity. Experiments show that LLMs fine-tuned with CTF-Instruct data achieve over a 2\% improvement on CTF-Code, and more than a 10\% performance boost on LiveCodeBench, validating the feasibility of enhancing LLMs' sensitivity to improve performance.
Ensuring the safety of large language models (LLMs) is critical for responsible deployment, yet existing evaluations often prioritize performance over identifying failure modes. We introduce Phare, a multilingual diagnostic framework to probe and evaluate LLM behavior across three critical dimensions: hallucination and reliability, social biases, and harmful content generation. Our evaluation of 17 state-of-the-art LLMs reveals patterns of systematic vulnerabilities across all safety dimensions, including sycophancy, prompt sensitivity, and stereotype reproduction. By highlighting these specific failure modes rather than simply ranking models, Phare provides researchers and practitioners with actionable insights to build more robust, aligned, and trustworthy language systems.
There can be many competing and contradictory explanations for a single model prediction, making it difficult to select which one to use. Current explanation evaluation frameworks measure quality by comparing against ideal "ground-truth" explanations, or by verifying model sensitivity to important inputs. We outline the limitations of these approaches, and propose three desirable principles to ground the future development of explanation evaluation strategies for local feature importance explanations. We propose a ground-truth Agnostic eXplanation Evaluation framework (AXE) for evaluating and comparing model explanations that satisfies these principles. Unlike prior approaches, AXE does not require access to ideal ground-truth explanations for comparison, or rely on model sensitivity - providing an independent measure of explanation quality. We verify AXE by comparing with baselines, and show how it can be used to detect explanation fairwashing. Our code is available at https://github.com/KaiRawal/Evaluating-Model-Explanations-without-Ground-Truth.
Background: There are many challenges and opportunities in the clinical deployment of AI tools in radiology. The current study describes a radiology software platform called NeoMedSys that can enable efficient deployment and refinements of AI models. We evaluated the feasibility and effectiveness of running NeoMedSys for three months in real-world clinical settings and focused on improvement performance of an in-house developed AI model (VIOLA-AI) designed for intracranial hemorrhage (ICH) detection. Methods: NeoMedSys integrates tools for deploying, testing, and optimizing AI models with a web-based medical image viewer, annotation system, and hospital-wide radiology information systems. A pragmatic investigation was deployed using clinical cases of patients presenting to the largest Emergency Department in Norway (site-1) with suspected traumatic brain injury (TBI) or patients with suspected stroke (site-2). We assessed ICH classification performance as VIOLA-AI encountered new data and underwent pre-planned model retraining. Performance metrics included sensitivity, specificity, accuracy, and the area under the receiver operating characteristic curve (AUC). Results: NeoMedSys facilitated iterative improvements in the AI model, significantly enhancing its diagnostic accuracy. Automated bleed detection and segmentation were reviewed in near real-time to facilitate re-training VIOLA-AI. The iterative refinement process yielded a marked improvement in classification sensitivity, rising to 90.3% (from 79.2%), and specificity that reached 89.3% (from 80.7%). The bleed detection ROC analysis for the entire sample demonstrated a high area-under-the-curve (AUC) of 0.949 (from 0.873). Model refinement stages were associated with notable gains, highlighting the value of real-time radiologist feedback.
Parametric differential equations of the form du/dt = f(u, x, t, p) are fundamental in science and engineering. While deep learning frameworks such as the Fourier Neural Operator (FNO) can efficiently approximate solutions, they struggle with inverse problems, sensitivity estimation (du/dp), and concept drift. We address these limitations by introducing a sensitivity-based regularization strategy, called Sensitivity-Constrained Fourier Neural Operators (SC-FNO). SC-FNO achieves high accuracy in predicting solution paths and consistently outperforms standard FNO and FNO with physics-informed regularization. It improves performance in parameter inversion tasks, scales to high-dimensional parameter spaces (tested with up to 82 parameters), and reduces both data and training requirements. These gains are achieved with a modest increase in training time (30% to 130% per epoch) and generalize across various types of differential equations and neural operators. Code and selected experiments are available at: https://github.com/AMBehroozi/SC_Neural_Operators
Mixture-of-Experts (MoE) models face deployment challenges due to their large parameter counts and computational demands. We explore quantization for MoE models and highlight two key insights: 1) linear blocks exhibit varying quantization sensitivity, and 2) divergent expert activation frequencies create heterogeneous computational characteristics. Based on these observations, we introduce MxMoE, a mixed-precision optimization framework for MoE models that considers both algorithmic and system perspectives. MxMoE navigates the design space defined by parameter sensitivity, expert activation dynamics, and hardware resources to derive efficient mixed-precision configurations. Additionally, MxMoE automatically generates optimized mixed-precision GroupGEMM kernels, enabling parallel execution of GEMMs with different precisions. Evaluations show that MxMoE outperforms existing methods, achieving 2.4 lower Wikitext-2 perplexity than GPTQ at 2.25-bit and delivering up to 3.4x speedup over full precision, as well as up to 29.4% speedup over uniform quantization at equivalent accuracy with 5-bit weight-activation quantization. Our code is available at https://github.com/cat538/MxMoE.
Machine learning (ML) models are often sensitive to carefully crafted yet seemingly unnoticeable perturbations. Such adversarial examples are considered to be a property of ML models, often associated with their black-box operation and sensitivity to features learned from data. This work examines the adversarial sensitivity of non-learned decision rules, and particularly of iterative optimizers. Our analysis is inspired by the recent developments in deep unfolding, which cast such optimizers as ML models. We show that non-learned iterative optimizers share the sensitivity to adversarial examples of ML models, and that attacking iterative optimizers effectively alters the optimization objective surface in a manner that modifies the minima sought. We then leverage the ability to cast iteration-limited optimizers as ML models to enhance robustness via adversarial training. For a class of proximal gradient optimizers, we rigorously prove how their learning affects adversarial sensitivity. We numerically back our findings, showing the vulnerability of various optimizers, as well as the robustness induced by unfolding and adversarial training.
Large Language Models (LLMs) are increasingly used to automate relevance judgments for information retrieval (IR) tasks, often demonstrating agreement with human labels that approaches inter-human agreement. To assess the robustness and reliability of LLM-based relevance judgments, we systematically investigate impact of prompt sensitivity on the task. We collected prompts for relevance assessment from 15 human experts and 15 LLMs across three tasks~ -- ~binary, graded, and pairwise~ -- ~yielding 90 prompts in total. After filtering out unusable prompts from three humans and three LLMs, we employed the remaining 72 prompts with three different LLMs as judges to label document/query pairs from two TREC Deep Learning Datasets (2020 and 2021). We compare LLM-generated labels with TREC official human labels using Cohen's $\kappa$ and pairwise agreement measures. In addition to investigating the impact of prompt variations on agreement with human labels, we compare human- and LLM-generated prompts and analyze differences among different LLMs as judges. We also compare human- and LLM-generated prompts with the standard UMBRELA prompt used for relevance assessment by Bing and TREC 2024 Retrieval Augmented Generation (RAG) Track. To support future research in LLM-based evaluation, we release all data and prompts at https://github.com/Narabzad/prompt-sensitivity-relevance-judgements/.
Radio Frequency Interference (RFI) is a known growing challenge for radio astronomy, intensified by increasing observatory sensitivity and prevalence of orbital RFI sources. Spiking Neural Networks (SNNs) offer a promising solution for real-time RFI detection by exploiting the time-varying nature of radio observation and neuron dynamics together. This work explores the inclusion of polarisation information in SNN-based RFI detection, using simulated data from the Hydrogen Epoch of Reionisation Array (HERA) instrument and provides power usage estimates for deploying SNN-based RFI detection on existing neuromorphic hardware. Preliminary results demonstrate state-of-the-art detection accuracy and highlight possible extensive energy-efficiency gains.
The epidemiological dynamics of Mycoplasma pneumoniae are characterized by complex and poorly understood multiannual cycles, posing challenges for forecasting. Using Bayesian methods to fit a seasonally forced transmission model to long-term surveillance data from Denmark (1958-1995, 2010-2025), we investigate the mechanisms driving recurrent outbreaks of M. pneumoniae. The period of the multiannual cycles (predominantly approx. 5 years in Denmark) are explained as a consequence of the interaction of two time-scales in the system, one intrinsic and one extrinsic (seasonal). While it provides an excellent fit to shorter time series (a few decades), we find that the deterministic model eventually settles into an annual cycle, failing to reproduce the observed 4-5-year periodicity long-term. Upon further analysis, the system is found to exhibit transient chaos and thus high sensitivity to stochasticity. We show that environmental (but not purely demographic) stochasticity can sustain the multi-year cycles via stochastic resonance. The disruptive effects of COVID-19 non-pharmaceutical interventions (NPIs) on M. pneumoniae circulation constitute a natural experiment on the effects of large perturbations. Consequently, the effects of NPIs are included in the model and medium-term predictions are explored. Our findings highlight the intrinsic sensitivity of M. pneumoniae dynamics to perturbations and interventions, underscoring the limitations of deterministic epidemic models for long-term prediction. More generally, our results emphasize the potential role of stochasticity as a driver of complex cycles across endemic and recurring pathogens.
Differentially private selection mechanisms offer strong privacy guarantees for queries aiming to identify the top-scoring element r from a finite set R, based on a dataset-dependent utility function. While selection queries are fundamental in data science, few mechanisms effectively ensure their privacy. Furthermore, most approaches rely on global sensitivity to achieve differential privacy (DP), which can introduce excessive noise and impair downstream inferences. To address this limitation, we propose the Smooth Noisy Max (SNM) mechanism, which leverages smooth sensitivity to yield provably tighter (upper bounds on) expected errors compared to global sensitivity-based methods. Empirical results demonstrate that SNM is more accurate than state-of-the-art differentially private selection methods in three applications: percentile selection, greedy decision trees, and random forests.
Lithium-ion batteries are widely used in transportation, energy storage, and consumer electronics, driving the need for reliable battery management systems (BMS) for state estimation and control. The Single Particle Model (SPM) balances computational efficiency and accuracy but faces challenges in parameter estimation due to numerous parameters. Current SPM models using parabolic approximation introduce intermediate variables and hard to do parameter grouping. This study presents a control-oriented SPM reformulation that employs parameter grouping and parabolic approximation to simplify model parameters while using average and surface lithium-ion concentrations as model output. By parameter grouping, the original 17 parameters were reduced to 9 grouped parameters. The reformulated model achieves a reduced-order ordinary differential equation form while maintaining mathematical accuracy equivalent to the pre-grouped discretized SPM. Through Sobol sensitivity analysis under various current profiles, the grouped parameters were reduced from 9 to 6 highly sensitive parameters. Results demonstrate that estimating these 6 parameters achieves comparable practical accuracy to estimating all 9 parameters, with faster convergence. This control-oriented SPM enhances BMS applications by facilitating state estimation and control while reducing parameter estimation requirements.