Research Papers

Derivative-Free Diffusion Manifold-Constrained Gradient for Unified XAI

Won Jun Kim, Hyungjin Chung, Jaemin Kim +3

Gradient-based methods are a prototypical family of explainability techniques, especially for image-based models. Nonetheless, they have several shortcomings in that they (1) require white-box access to models, (2) are vulnerable to adversarial attacks, and (3) produce attributions that lie off the image manifold, leading to explanations that are not actually faithful to the model and do not align well with human perception. To overcome these challenges, we introduce Derivative-Free Diffusion Manifold-Constrainted Gradients (FreeMCG), a novel method that serves as an improved basis for explainability of a given neural network than the traditional gradient. Specifically, by leveraging ensemble Kalman filters and diffusion models, we derive a derivative-free approximation of the model's gradient projected onto the data manifold, requiring access only to the model's outputs. We demonstrate the effectiveness of FreeMCG by applying it to both counterfactual generation and feature attribution, which have traditionally been treated as distinct tasks. Through comprehensive evaluation on both tasks, counterfactual explanation and feature attribution, we show that our method yields state-of-the-art results while preserving the essential properties expected of XAI tools.

MEGL: Multimodal Explanation-Guided Learning

Yifei Zhang, Tianxu Jiang, Bo Pan +3

Explaining the decision-making processes of Artificial Intelligence (AI) models is crucial for addressing their "black box" nature, particularly in tasks like image classification. Traditional eXplainable AI (XAI) methods typically rely on unimodal explanations, either visual or textual, each with inherent limitations. Visual explanations highlight key regions but often lack rationale, while textual explanations provide context without spatial grounding. Further, both explanation types can be inconsistent or incomplete, limiting their reliability. To address these challenges, we propose a novel Multimodal Explanation-Guided Learning (MEGL) framework that leverages both visual and textual explanations to enhance model interpretability and improve classification performance. Our Saliency-Driven Textual Grounding (SDTG) approach integrates spatial information from visual explanations into textual rationales, providing spatially grounded and contextually rich explanations. Additionally, we introduce Textual Supervision on Visual Explanations to align visual explanations with textual rationales, even in cases where ground truth visual annotations are missing. A Visual Explanation Distribution Consistency loss further reinforces visual coherence by aligning the generated visual explanations with dataset-level patterns, enabling the model to effectively learn from incomplete multimodal supervision. We validate MEGL on two new datasets, Object-ME and Action-ME, for image classification with multimodal explanations. Experimental results demonstrate that MEGL outperforms previous approaches in prediction accuracy and explanation quality across both visual and textual domains. Our code will be made available upon the acceptance of the paper.

Exploring the Requirements of Clinicians for Explainable AI Decision Support Systems in Intensive Care

Jeffrey N. Clark, Matthew Wragg, Emily Nielsen +7

There is a growing need to understand how digital systems can support clinical decision-making, particularly as artificial intelligence (AI) models become increasingly complex and less human-interpretable. This complexity raises concerns about trustworthiness, impacting safe and effective adoption of such technologies. Improved understanding of decision-making processes and requirements for explanations coming from decision support tools is a vital component in providing effective explainable solutions. This is particularly relevant in the data-intensive, fast-paced environments of intensive care units (ICUs). To explore these issues, group interviews were conducted with seven ICU clinicians, representing various roles and experience levels. Thematic analysis revealed three core themes: (T1) ICU decision-making relies on a wide range of factors, (T2) the complexity of patient state is challenging for shared decision-making, and (T3) requirements and capabilities of AI decision support systems. We include design recommendations from clinical input, providing insights to inform future AI systems for intensive care.

Neural additive model (NAM) is a recently proposed explainable artificial intelligence (XAI) method that utilizes neural network-based architectures. Given the advantages of neural networks, NAMs provide intuitive explanations for their predictions with high model performance. In this paper, we analyze a critical yet overlooked phenomenon: NAMs often produce inconsistent explanations, even when using the same architecture and dataset. Traditionally, such inconsistencies have been viewed as issues to be resolved. However, we argue instead that these inconsistencies can provide valuable explanations within the given data model. Through a simple theoretical framework, we demonstrate that these inconsistencies are not mere artifacts but emerge naturally in datasets with multiple important features. To effectively leverage this information, we introduce a novel framework, Bayesian Neural Additive Model (BayesNAM), which integrates Bayesian neural networks and feature dropout, with theoretical proof demonstrating that feature dropout effectively captures model inconsistencies. Our experiments demonstrate that BayesNAM effectively reveals potential problems such as insufficient data or structural limitations of the model, providing more reliable explanations and potential remedies.

Causal Responsibility Attribution for Human-AI Collaboration

Yahang Qi, Bernhard Schölkopf, Zhijing Jin

As Artificial Intelligence (AI) systems increasingly influence decision-making across various fields, the need to attribute responsibility for undesirable outcomes has become essential, though complicated by the complex interplay between humans and AI. Existing attribution methods based on actual causality and Shapley values tend to disproportionately blame agents who contribute more to an outcome and rely on real-world measures of blameworthiness that may misalign with responsible AI standards. This paper presents a causal framework using Structural Causal Models (SCMs) to systematically attribute responsibility in human-AI systems, measuring overall blameworthiness while employing counterfactual reasoning to account for agents' expected epistemic levels. Two case studies illustrate the framework's adaptability in diverse human-AI collaboration scenarios.

Explainable AI (XAI) aims to support appropriate human-AI reliance by increasing the interpretability of complex model decisions. Despite the proliferation of proposed methods, there is mixed evidence surrounding the effects of different styles of XAI explanations on human-AI reliance. Interpreting these conflicting findings requires an understanding of the individual and combined qualities of different explanation styles that influence appropriate and inappropriate human-AI reliance, and the role of interpretability in this interaction. In this study, we investigate the influences of feature-based, example-based, and combined feature- and example-based XAI methods on human-AI reliance through a two-part experimental study with 274 participants comparing these explanation style conditions. Our findings suggest differences between feature-based and example-based explanation styles beyond interpretability that affect human-AI reliance patterns across differences in individual performance and task complexity. Our work highlights the importance of adapting explanations to their specific users and context over maximising broad interpretability.

Differentiating between Intestinal Tuberculosis (ITB) and Crohn's Disease (CD) poses a significant clinical challenge due to their similar symptoms, clinical presentations, and imaging features. This study leverages Computed Tomography Enterography (CTE) scans, deep learning, and traditional computer vision to address this diagnostic dilemma. A consensus among radiologists from renowned institutions has recognized the visceral-to-subcutaneous fat (VF/SF) ratio as a surrogate biomarker for differentiating between ITB and CD. Previously done manually, we propose a novel 2D image computer vision algorithm for auto-segmenting subcutaneous fat to automate this ratio calculation, enhancing diagnostic efficiency and objectivity. As a benchmark, we compare the results to those obtained using the TotalSegmentator tool, a popular deep learning-based software for automatic segmentation of anatomical structures, and manual calculations by radiologists. We also demonstrated the performance on 3D CT volumes using a slicing method and provided a benchmark comparison of the algorithm with the TotalSegmentator tool. Additionally, we propose a scoring approach to integrate scores from radiological features, such as the fat ratio and pulmonary TB probability, into a single score for diagnosis. We trained a ResNet10 model on a dataset of CTE scans with samples from ITB, CD, and normal patients, achieving an accuracy of 75%. To enhance interpretability and gain clinical trust, we integrated the explainable AI technique Grad-CAM with ResNet10 to explain the model's predictions. Due to the small dataset size (100 total cases), the feature-based scoring system is considered more reliable and trusted by radiologists compared to the deep learning model for disease diagnosis.

Activation maximization (AM) strives to generate optimal input stimuli, revealing features that trigger high responses in trained deep neural networks. AM is an important method of explainable AI. We demonstrate that AM fails to produce optimal input stimuli for simple functions containing ReLUs or Leaky ReLUs, casting doubt on the practical usefulness of AM and the visual interpretation of the generated images. This paper proposes a solution based on using Leaky ReLUs with a high negative slope in the backward pass while keeping the original, usually zero, slope in the forward pass. The approach significantly increases the maxima found by AM. The resulting ProxyGrad algorithm implements a novel optimization technique for neural networks that employs a secondary network as a proxy for gradient computation. This proxy network is designed to have a simpler loss landscape with fewer local maxima than the original network. Our chosen proxy network is an identical copy of the original network, including its weights, with distinct negative slopes in the Leaky ReLUs. Moreover, we show that ProxyGrad can be used to train the weights of Convolutional Neural Networks for classification such that, on some of the tested benchmarks, they outperform traditional networks.

The impressive performance of deep learning models, particularly Convolutional Neural Networks (CNNs), is often hindered by their lack of interpretability, rendering them "black boxes." This opacity raises concerns in critical areas like healthcare, finance, and autonomous systems, where trust and accountability are crucial. This paper introduces the QIXAI Framework (Quantum-Inspired Explainable AI), a novel approach for enhancing neural network interpretability through quantum-inspired techniques. By utilizing principles from quantum mechanics, such as Hilbert spaces, superposition, entanglement, and eigenvalue decomposition, the QIXAI framework reveals how different layers of neural networks process and combine features to make decisions. We critically assess model-agnostic methods like SHAP and LIME, as well as techniques like Layer-wise Relevance Propagation (LRP), highlighting their limitations in providing a comprehensive view of neural network operations. The QIXAI framework overcomes these limitations by offering deeper insights into feature importance, inter-layer dependencies, and information propagation. A CNN for malaria parasite detection is used as a case study to demonstrate how quantum-inspired methods like Singular Value Decomposition (SVD), Principal Component Analysis (PCA), and Mutual Information (MI) provide interpretable explanations of model behavior. Additionally, we explore the extension of QIXAI to other architectures, including Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, Transformers, and Natural Language Processing (NLP) models, and its application to generative models and time-series analysis. The framework applies to both quantum and classical systems, demonstrating its potential to improve interpretability and transparency across a range of models, advancing the broader goal of developing trustworthy AI systems.

How Performance Pressure Influences AI-Assisted Decision Making

Nikita Haduong, Noah A. Smith

Many domains now employ AI-based decision-making aids, and although the potential for AI systems to assist with decision making is much discussed, human-AI collaboration often underperforms due to factors such as (mis)trust in the AI system and beliefs about AI being incapable of completing subjective tasks. One potential tool for influencing human decision making is performance pressure, which hasn't been much studied in interaction with human-AI decision making. In this work, we examine how pressure and explainable AI (XAI) techniques interact with AI advice-taking behavior. Using an inherently low-stakes task (spam review classification), we demonstrate effective and simple methods to apply pressure and influence human AI advice-taking behavior by manipulating financial incentives and imposing time limits. Our results show complex interaction effects, with different combinations of pressure and XAI techniques either improving or worsening AI advice taking behavior. We conclude by discussing the implications of these interactions, strategies to effectively use pressure, and encourage future research to incorporate pressure analysis.

In this paper, we explore how performers' embodied interactions with a Neural Audio Synthesis model allow the exploration of the latent space of such a model, mediated through movements sensed by e-textiles. We provide background and context for the performance, highlighting the potential of embodied practices to contribute to developing explainable AI systems. By integrating various artistic domains with explainable AI principles, our interdisciplinary exploration contributes to the discourse on art, embodiment, and AI, offering insights into intuitive approaches found through bodily expression.

Deep learning (DL) has significantly improved automatic modulation classification (AMC) by leveraging neural networks as the feature extractor.However, as the DL-based AMC becomes increasingly widespread, it is faced with the severe secure issue from various adversarial attacks. Existing defense methods often suffer from the high computational cost, intractable parameter tuning, and insufficient robustness.This paper proposes an eXplainable artificial intelligence (XAI) defense approach, which uncovers the negative information caused by the adversarial attack through measuring the importance of input features based on the SHapley Additive exPlanations (SHAP).By properly removing the negative information in adversarial samples and then fine-tuning(FT) the model, the impact of the attacks on the classification result can be mitigated.Experimental results demonstrate that the proposed SHAP-FT improves the classification performance of the model by 15%-20% under different attack levels,which not only enhances model robustness against various attack levels but also reduces the resource consumption, validating its effectiveness in safeguarding communication networks.