Research Papers

Beyond the Veil of Similarity: Quantifying Semantic Continuity in Explainable AI

Qi Huang, Emanuele Mezzi, Osman Mutlu +5

We introduce a novel metric for measuring semantic continuity in Explainable AI methods and machine learning models. We posit that for models to be truly interpretable and trustworthy, similar inputs should yield similar explanations, reflecting a consistent semantic understanding. By leveraging XAI techniques, we assess semantic continuity in the task of image recognition. We conduct experiments to observe how incremental changes in input affect the explanations provided by different XAI methods. Through this approach, we aim to evaluate the models' capability to generalize and abstract semantic concepts accurately and to evaluate different XAI methods in correctly capturing the model behaviour. This paper contributes to the broader discourse on AI interpretability by proposing a quantitative measure for semantic continuity for XAI methods, offering insights into the models' and explainers' internal reasoning processes, and promoting more reliable and transparent AI systems.

Models like Delphi have been able to label ethical dilemmas as moral or immoral with astonishing accuracy. This paper challenges accuracy as a holistic metric for ethics modeling by identifying issues with translating moral dilemmas into text-based input. It demonstrates these issues with contrast sets that substantially reduce the performance of classifiers trained on the dataset Moral Stories. Ultimately, we obtain concrete estimates for how much specific forms of data misrepresentation harm classifier accuracy. Specifically, label-changing tweaks to the descriptive content of a situation (as small as 3-5 words) can reduce classifier accuracy to as low as 51%, almost half the initial accuracy of 99.8%. Associating situations with a misleading social norm lowers accuracy to 98.8%, while adding textual bias (i.e. an implication that a situation already fits a certain label) lowers accuracy to 77%. These results suggest not only that many ethics models have substantially overfit, but that several precautions are required to ensure that input accurately captures a moral dilemma. This paper recommends re-examining the structure of a social norm, training models to ask for context with defeasible reasoning, and filtering input for textual bias. Doing so not only gives us the first concrete estimates of the average cost to accuracy of misrepresenting ethics data, but gives researchers practical tips for considering these estimates in research.

Our society collects data on people for a wide range of applications, from building a census for policy evaluation to running meaningful clinical trials. To collect data, we typically sample individuals with the goal of accurately representing a population of interest. However, current sampling processes often collect data opportunistically from data sources, which can lead to datasets that are biased and not representative, i.e., the collected dataset does not accurately reflect the distribution of demographics of the true population. This is a concern because subgroups within the population can be under- or over-represented in a dataset, which may harm generalizability and lead to an unequal distribution of benefits and harms from downstream tasks that use such datasets (e.g., algorithmic bias in medical decision-making algorithms). In this paper, we assess the relationship between dataset representativeness and group-fairness of classifiers trained on that dataset. We demonstrate that there is a natural tension between dataset representativeness and classifier fairness; empirically we observe that training datasets with better representativeness can frequently result in classifiers with higher rates of unfairness. We provide some intuition as to why this occurs via a set of theoretical results in the case of univariate classifiers. We also find that over-sampling underrepresented groups can result in classifiers which exhibit greater bias to those groups. Lastly, we observe that fairness-aware sampling strategies (i.e., those which are specifically designed to select data with high downstream fairness) will often over-sample members of majority groups. These results demonstrate that the relationship between dataset representativeness and downstream classifier fairness is complex; balancing these two quantities requires special care from both model- and dataset-designers.

This study addresses gender bias in image generation models using Reinforcement Learning from Artificial Intelligence Feedback (RLAIF) with a novel Denoising Diffusion Policy Optimization (DDPO) pipeline. By employing a pretrained stable diffusion model and a highly accurate gender classification Transformer, the research introduces two reward functions: Rshift for shifting gender imbalances, and Rbalance for achieving and maintaining gender balance. Experiments demonstrate the effectiveness of this approach in mitigating bias without compromising image quality or requiring additional data or prompt modifications. While focusing on gender bias, this work establishes a foundation for addressing various forms of bias in AI systems, emphasizing the need for responsible AI development. Future research directions include extending the methodology to other bias types, enhancing the RLAIF pipeline's robustness, and exploring multi-prompt fine-tuning to further advance fairness and inclusivity in AI.

In Explainable AI, rule extraction translates model knowledge into logical rules, such as IF-THEN statements, crucial for understanding patterns learned by black-box models. This could significantly aid in fields like disease diagnosis, disease progression estimation, or drug discovery. However, such application domains often contain imbalanced data, with the class of interest underrepresented. Existing methods inevitably compromise the performance of rules for the minor class to maximise the overall performance. As the first attempt in this field, we propose a model-agnostic approach for extracting rules from specific subgroups of data, featuring automatic rule generation for numerical features. This method enhances the regional explainability of machine learning models and offers wider applicability compared to existing methods. We additionally introduce a new method for selecting features to compose rules, reducing computational costs in high-dimensional spaces. Experiments across various datasets and models demonstrate the effectiveness of our methods.

Hardware-Aware Neural Dropout Search for Reliable Uncertainty Prediction on FPGA

Zehuan Zhang, Hongxiang Fan, Hao Mark Chen +2

The increasing deployment of artificial intelligence (AI) for critical decision-making amplifies the necessity for trustworthy AI, where uncertainty estimation plays a pivotal role in ensuring trustworthiness. Dropout-based Bayesian Neural Networks (BayesNNs) are prominent in this field, offering reliable uncertainty estimates. Despite their effectiveness, existing dropout-based BayesNNs typically employ a uniform dropout design across different layers, leading to suboptimal performance. Moreover, as diverse applications require tailored dropout strategies for optimal performance, manually optimizing dropout configurations for various applications is both error-prone and labor-intensive. To address these challenges, this paper proposes a novel neural dropout search framework that automatically optimizes both the dropout-based BayesNNs and their hardware implementations on FPGA. We leverage one-shot supernet training with an evolutionary algorithm for efficient dropout optimization. A layer-wise dropout search space is introduced to enable the automatic design of dropout-based BayesNNs with heterogeneous dropout configurations. Extensive experiments demonstrate that our proposed framework can effectively find design configurations on the Pareto frontier. Compared to manually-designed dropout-based BayesNNs on GPU, our search approach produces FPGA designs that can achieve up to 33X higher energy efficiency. Compared to state-of-the-art FPGA designs of BayesNN, the solutions from our approach can achieve higher algorithmic performance and energy efficiency.

Magnetic resonance imaging (MRI) has become a crucial tool in the diagnosis and staging of prostate cancer, owing to its superior tissue contrast. However, it also creates large volumes of data that must be assessed by trained experts, a time-consuming and laborious task. This has prompted the development of machine learning tools for the automation of Prostate cancer (PCa) risk classification based on multiple MRI modalities (T2W, ADC, and high-b-value DWI). Understanding and interpreting the predictions made by the models, however, remains a challenge. We analyze Random Forests (RF) and Support Vector Machines (SVM), for two complementary datasets, the public Prostate-X dataset, and an in-house, mostly early-stage PCa dataset to elucidate the contributions made by first-order statistical features, Haralick texture features, and local binary patterns to the classification. Using correlation analysis and Shapley impact scores, we find that many of the features typically used are strongly correlated, and that the majority of features have negligible impact on the classification. We identify a small set of features that determine the classification outcome, which may aid the development of explainable AI approaches.

Explaning with trees: interpreting CNNs using hierarchies

Caroline Mazini Rodrigues, Nicolas Boutry, Laurent Najman

Challenges persist in providing interpretable explanations for neural network reasoning in explainable AI (xAI). Existing methods like Integrated Gradients produce noisy maps, and LIME, while intuitive, may deviate from the model's reasoning. We introduce a framework that uses hierarchical segmentation techniques for faithful and interpretable explanations of Convolutional Neural Networks (CNNs). Our method constructs model-based hierarchical segmentations that maintain the model's reasoning fidelity and allows both human-centric and model-centric segmentation. This approach offers multiscale explanations, aiding bias identification and enhancing understanding of neural network decision-making. Experiments show that our framework, xAiTrees, delivers highly interpretable and faithful model explanations, not only surpassing traditional xAI methods but shedding new light on a novel approach to enhancing xAI interpretability.

Explainable AI Planning (XAIP) aims to develop AI agents that can effectively explain their decisions and actions to human users, fostering trust and facilitating human-AI collaboration. A key challenge in XAIP is model reconciliation, which seeks to align the mental models of AI agents and humans. While existing approaches often assume a known and deterministic human model, this simplification may not capture the complexities and uncertainties of real-world interactions. In this paper, we propose a novel framework that enables AI agents to learn and update a probabilistic human model through argumentation-based dialogues. Our approach incorporates trust-based and certainty-based update mechanisms, allowing the agent to refine its understanding of the human's mental state based on the human's expressed trust in the agent's arguments and certainty in their own arguments. We employ a probability weighting function inspired by prospect theory to capture the relationship between trust and perceived probability, and use a Bayesian approach to update the agent's probability distribution over possible human models. We conduct a human-subject study to empirically evaluate the effectiveness of our approach in an argumentation scenario, demonstrating its ability to capture the dynamics of human belief formation and adaptation.

Primary care providers are vital for initial triage and referrals to specialty care. In glaucoma, asymptomatic and fast progression can lead to vision loss, necessitating timely referrals to specialists. However, primary eye care providers may not identify urgent cases, potentially delaying care. Artificial Intelligence (AI) offering explanations could enhance their referral decisions. We investigate how various AI explanations help providers distinguish between patients needing immediate or non-urgent specialist referrals. We built explainable AI algorithms to predict glaucoma surgery needs from routine eyecare data as a proxy for identifying high-risk patients. We incorporated intrinsic and post-hoc explainability and conducted an online study with optometrists to assess human-AI team performance, measuring referral accuracy and analyzing interactions with AI, including agreement rates, task time, and user experience perceptions. AI support enhanced referral accuracy among 87 participants (59.9%/50.8% with/without AI), though Human-AI teams underperformed compared to AI alone. Participants believed they included AI advice more when using the intrinsic model, and perceived it more useful and promising. Without explanations, deviations from AI recommendations increased. AI support did not increase workload, confidence, and trust, but reduced challenges. On a separate test set, our black-box and intrinsic models achieved an accuracy of 77% and 71%, respectively, in predicting surgical outcomes. We identify opportunities of human-AI teaming for glaucoma management in primary eye care, noting that while AI enhances referral accuracy, it also shows a performance gap compared to AI alone, even with explanations. Human involvement remains essential in medical decision making, underscoring the need for future research to optimize collaboration, ensuring positive experiences and safe AI use.

Heatmaps have been instrumental in helping understand deep network decisions, and are a common approach for Explainable AI (XAI). While significant progress has been made in enhancing the informativeness and accessibility of heatmaps, heatmap analysis is typically very subjective and limited to domain experts. As such, developing automatic, scalable, and numerical analysis methods to make heatmap-based XAI more objective, end-user friendly, and cost-effective is vital. In addition, there is a need for comprehensive evaluation metrics to assess heatmap quality at a granular level.

The increasing complexity of Artificial Intelligence models poses challenges to interpretability, particularly in the healthcare sector. This study investigates the impact of deep learning model complexity and Explainable AI (XAI) efficacy, utilizing four ResNet architectures (ResNet-18, 34, 50, 101). Through methodical experimentation on 4,369 lung X-ray images of COVID-19-infected and healthy patients, the research evaluates models' classification performance and the relevance of corresponding XAI explanations with respect to the ground-truth disease masks. Results indicate that the increase in model complexity is associated with a decrease in classification accuracy and AUC-ROC scores (ResNet-18: 98.4%, 0.997; ResNet-101: 95.9%, 0.988). Notably, in eleven out of twelve statistical tests performed, no statistically significant differences occurred between XAI quantitative metrics - Relevance Rank Accuracy and the proposed Positive Attribution Ratio - across trained models. These results suggest that increased model complexity does not consistently lead to higher performance or relevance of explanations for models' decision-making processes.