Research Papers

FitCF: A Framework for Automatic Feature Importance-guided Counterfactual Example Generation

Qianli Wang, Nils Feldhus, Simon Ostermann +3

Counterfactual examples are widely used in natural language processing (NLP) as valuable data to improve models, and in explainable artificial intelligence (XAI) to understand model behavior. The automated generation of counterfactual examples remains a challenging task even for large language models (LLMs), despite their impressive performance on many tasks. In this paper, we first introduce ZeroCF, a faithful approach for leveraging important words derived from feature attribution methods to generate counterfactual examples in a zero-shot setting. Second, we present a new framework, FitCF, which further verifies aforementioned counterfactuals by label flip verification and then inserts them as demonstrations for few-shot prompting, outperforming two state-of-the-art baselines. Through ablation studies, we identify the importance of each of FitCF's core components in improving the quality of counterfactuals, as assessed through flip rate, perplexity, and similarity measures. Furthermore, we show the effectiveness of LIME and Integrated Gradients as backbone attribution methods for FitCF and find that the number of demonstrations has the largest effect on performance. Finally, we reveal a strong correlation between the faithfulness of feature attribution scores and the quality of generated counterfactuals, which we hope will serve as an important finding for future research in this direction.

Responsible AI principles provide ethical guidelines for developing AI systems, yet their practical implementation in software engineering lacks thorough investigation. Therefore, this study explores the practices and challenges faced by software practitioners in aligning with these principles. Through semi-structured interviews with 25 practitioners, we investigated their methods, concerns, and strategies for addressing Responsible AI in software development. Our findings reveal that while practitioners frequently address fairness, inclusiveness, and reliability, principles such as transparency and accountability receive comparatively less attention in their practices. This scenario highlights gaps in current strategies and the need for more comprehensive frameworks to fully operationalize Responsible AI principles in software engineering.

The lack of ground truth explanation labels is a fundamental challenge for quantitative evaluation in explainable artificial intelligence (XAI). This challenge becomes especially problematic when evaluation methods have numerous hyperparameters that must be specified by the user, as there is no ground truth to determine an optimal hyperparameter selection. It is typically not feasible to do an exhaustive search of hyperparameters so researchers typically make a normative choice based on similar studies in the literature, which provides great flexibility for the user. In this work, we illustrate how this flexibility can be exploited to manipulate the evaluation outcome. We frame this manipulation as an adversarial attack on the evaluation where seemingly innocent changes in hyperparameter setting significantly influence the evaluation outcome. We demonstrate the effectiveness of our manipulation across several datasets with large changes in evaluation outcomes across several explanation methods and models. Lastly, we propose a mitigation strategy based on ranking across hyperparameters that aims to provide robustness towards such manipulation. This work highlights the difficulty of conducting reliable XAI evaluation and emphasizes the importance of a holistic and transparent approach to evaluation in XAI.

Algorithmic bias continues to be a key concern of learning analytics. We study the statistical properties of the Absolute Between-ROC Area (ABROCA) metric. This fairness measure quantifies group-level differences in classifier performance through the absolute difference in ROC curves. ABROCA is particularly useful for detecting nuanced performance differences even when overall Area Under the ROC Curve (AUC) values are similar. We sample ABROCA under various conditions, including varying AUC differences and class distributions. We find that ABROCA distributions exhibit high skewness dependent on sample sizes, AUC differences, and class imbalance. When assessing whether a classifier is biased, this skewness inflates ABROCA values by chance, even when data is drawn (by simulation) from populations with equivalent ROC curves. These findings suggest that ABROCA requires careful interpretation given its distributional properties, especially when used to assess the degree of bias and when classes are imbalanced.

As AI systems become integral to critical operations across industries and services, ensuring their reliability and safety is essential. We offer a framework that integrates established reliability and resilience engineering principles into AI systems. By applying traditional metrics such as failure rate and Mean Time Between Failures (MTBF) along with resilience engineering and human reliability analysis, we propose an integrate framework to manage AI system performance, and prevent or efficiently recover from failures. Our work adapts classical engineering methods to AI systems and outlines a research agenda for future technical studies. We apply our framework to a real-world AI system, using system status data from platforms such as openAI, to demonstrate its practical applicability. This framework aligns with emerging global standards and regulatory frameworks, providing a methodology to enhance the trustworthiness of AI systems. Our aim is to guide policy, regulation, and the development of reliable, safe, and adaptable AI technologies capable of consistent performance in real-world environments.

Transformer-based models have achieved state-of-the-art performance in various computer vision tasks, including image and video analysis. However, Transformer's complex architecture and black-box nature pose challenges for explainability, a crucial aspect for real-world applications and scientific inquiry. Current Explainable AI (XAI) methods can only provide one-dimensional feature importance, either spatial or temporal explanation, with significant computational complexity. This paper introduces STAA (Spatio-Temporal Attention Attribution), an XAI method for interpreting video Transformer models. Differ from traditional methods that separately apply image XAI techniques for spatial features or segment contribution analysis for temporal aspects, STAA offers both spatial and temporal information simultaneously from attention values in Transformers. The study utilizes the Kinetics-400 dataset, a benchmark collection of 400 human action classes used for action recognition research. We introduce metrics to quantify explanations. We also apply optimization to enhance STAA's raw output. By implementing dynamic thresholding and attention focusing mechanisms, we improve the signal-to-noise ratio in our explanations, resulting in more precise visualizations and better evaluation results. In terms of computational overhead, our method requires less than 3\% of the computational resources of traditional XAI methods, making it suitable for real-time video XAI analysis applications. STAA contributes to the growing field of XAI by offering a method for researchers and practitioners to analyze Transformer models.

Deep Learning methods are notorious for relying on extensive labeled datasets to train and assess their performance. This can cause difficulties in practical situations where models should be trained for new applications for which very little data is available. While few-shot learning algorithms can address the first problem, they still lack sufficient explanations for the results. This research presents a workflow that tackles both challenges by proposing an explainable few-shot learning workflow for detecting invasive and exotic tree species in the Atlantic Forest of Brazil using Unmanned Aerial Vehicle (UAV) images. By integrating a Siamese network with explainable AI (XAI), the workflow enables the classification of tree species with minimal labeled data while providing visual, case-based explanations for the predictions. Results demonstrate the effectiveness of the proposed workflow in identifying new tree species, even in data-scarce conditions. With a lightweight backbone, e.g., MobileNet, it achieves a F1-score of 0.86 in 3-shot learning, outperforming a shallow CNN. A set of explanation metrics, i.e., correctness, continuity, and contrastivity, accompanied by visual cases, provide further insights about the prediction results. This approach opens new avenues for using AI and UAVs in forest management and biodiversity conservation, particularly concerning rare or under-studied species.

Audits contribute to the trustworthiness of Learning Analytics (LA) systems that integrate Artificial Intelligence (AI) and may be legally required in the future. We argue that the efficacy of an audit depends on the auditability of the audited system. Therefore, systems need to be designed with auditability in mind. We present a framework for assessing the auditability of AI-integrating systems that consists of three parts: (1) Verifiable claims about the validity, utility and ethics of the system, (2) Evidence on subjects (data, models or the system) in different types (documentation, raw sources and logs) to back or refute claims, (3) Evidence must be accessible to auditors via technical means (APIs, monitoring tools, explainable AI, etc.). We apply the framework to assess the auditability of Moodle's dropout prediction system and a prototype AI-based LA. We find that Moodle's auditability is limited by incomplete documentation, insufficient monitoring capabilities and a lack of available test data. The framework supports assessing the auditability of AI-based LA systems in use and improves the design of auditable systems and thus of audits.

As AI-enhanced academic search systems become increasingly popular among researchers, investigating their AI transparency is crucial to ensure trust in the search outcomes, as well as the reliability and integrity of scholarly work. This study employs a qualitative content analysis approach to examine the websites of a sample of 10 AI-enhanced academic search systems identified through university library guides. The assessed level of transparency varies across these systems: five provide detailed information about their mechanisms, three offer partial information, and two provide little to no information. These findings indicate that the academic community is recommending and using tools with opaque functionalities, raising concerns about research integrity, including issues of reproducibility and researcher responsibility.

Numerous fairness metrics have been proposed and employed by artificial intelligence (AI) experts to quantitatively measure bias and define fairness in AI models. Recognizing the need to accommodate stakeholders' diverse fairness understandings, efforts are underway to solicit their input. However, conveying AI fairness metrics to stakeholders without AI expertise, capturing their personal preferences, and seeking a collective consensus remain challenging and underexplored. To bridge this gap, we propose a new framework, EARN Fairness, which facilitates collective metric decisions among stakeholders without requiring AI expertise. The framework features an adaptable interactive system and a stakeholder-centered EARN Fairness process to Explain fairness metrics, Ask stakeholders' personal metric preferences, Review metrics collectively, and Negotiate a consensus on metric selection. To gather empirical results, we applied the framework to a credit rating scenario and conducted a user study involving 18 decision subjects without AI knowledge. We identify their personal metric preferences and their acceptable level of unfairness in individual sessions. Subsequently, we uncovered how they reached metric consensus in team sessions. Our work shows that the EARN Fairness framework enables stakeholders to express personal preferences and reach consensus, providing practical guidance for implementing human-centered AI fairness in high-risk contexts. Through this approach, we aim to harmonize fairness expectations of diverse stakeholders, fostering more equitable and inclusive AI fairness.

Biased information (recently termed bisinformation) continues to be taught in medical curricula, often long after having been debunked. In this paper, we introduce BRICC, a firstin-class initiative that seeks to mitigate medical bisinformation using machine learning to systematically identify and flag text with potential biases, for subsequent review in an expert-in-the-loop fashion, thus greatly accelerating an otherwise labor-intensive process. A gold-standard BRICC dataset was developed throughout several years, and contains over 12K pages of instructional materials. Medical experts meticulously annotated these documents for bias according to comprehensive coding guidelines, emphasizing gender, sex, age, geography, ethnicity, and race. Using this labeled dataset, we trained, validated, and tested medical bias classifiers. We test three classifier approaches: a binary type-specific classifier, a general bias classifier; an ensemble combining bias type-specific classifiers independently-trained; and a multitask learning (MTL) model tasked with predicting both general and type-specific biases. While MTL led to some improvement on race bias detection in terms of F1-score, it did not outperform binary classifiers trained specifically on each task. On general bias detection, the binary classifier achieves up to 0.923 of AUC, a 27.8% improvement over the baseline. This work lays the foundations for debiasing medical curricula by exploring a novel dataset and evaluating different training model strategies. Hence, it offers new pathways for more nuanced and effective mitigation of bisinformation.

Recent regulatory initiatives like the European AI Act and relevant voices in the Machine Learning (ML) community stress the need to describe datasets along several key dimensions for trustworthy AI, such as the provenance processes and social concerns. However, this information is typically presented as unstructured text in accompanying documentation, hampering their automated analysis and processing. In this work, we explore using large language models (LLM) and a set of prompting strategies to automatically extract these dimensions from documents and enrich the dataset description with them. Our approach could aid data publishers and practitioners in creating machine-readable documentation to improve the discoverability of their datasets, assess their compliance with current AI regulations, and improve the overall quality of ML models trained on them. In this paper, we evaluate the approach on 12 scientific dataset papers published in two scientific journals (Nature's Scientific Data and Elsevier's Data in Brief) using two different LLMs (GPT3.5 and Flan-UL2). Results show good accuracy with our prompt extraction strategies. Concrete results vary depending on the dimensions, but overall, GPT3.5 shows slightly better accuracy (81,21%) than FLAN-UL2 (69,13%) although it is more prone to hallucinations. We have released an open-source tool implementing our approach and a replication package, including the experiments' code and results, in an open-source repository.