Research Papers

The paper introduces concepts of fairness and explainability (XAI) in artificial intelligence, oriented to solve a sophisticated business problems. For fairness, the authors discuss the bias-inducing specifics, as well as relevant mitigation methods, concluding with a set of recipes for introducing fairness in data-driven organizations. Additionally, for XAI, the authors audit specific algorithms paired with demonstrational business use-cases, discuss a plethora of techniques of explanations quality quantification and provide an overview of future research avenues.

Binarized Neural Networks (BNNs) have the potential to revolutionize the way that deep learning is carried out in edge computing platforms. However, the effectiveness of interpretability methods on these networks has not been assessed. In this paper, we compare the performance of several widely used saliency map-based interpretabilty techniques (Gradient, SmoothGrad and GradCAM), when applied to Binarized or Full Precision Neural Networks (FPNNs). We found that the basic Gradient method produces very similar-looking maps for both types of network. However, SmoothGrad produces significantly noisier maps for BNNs. GradCAM also produces saliency maps which differ between network types, with some of the BNNs having seemingly nonsensical explanations. We comment on possible reasons for these differences in explanations and present it as an example of why interpretability techniques should be tested on a wider range of network types.

Explainability techniques for Graph Neural Networks still have a long way to go compared to explanations available for both neural and decision decision tree-based models trained on tabular data. Using a task that straddles both graphs and tabular data, namely Entity Matching, we comment on key aspects of explainability that are missing in GNN model explanations.

Despite the high accuracy offered by state-of-the-art deep natural-language models (e.g. LSTM, BERT), their application in real-life settings is still widely limited, as they behave like a black-box to the end-user. Hence, explainability is rapidly becoming a fundamental requirement of future-generation data-driven systems based on deep-learning approaches. Several attempts to fulfill the existing gap between accuracy and interpretability have been done. However, robust and specialized xAI (Explainable Artificial Intelligence) solutions tailored to deep natural-language models are still missing. We propose a new framework, named T-EBAnO, which provides innovative prediction-local and class-based model-global explanation strategies tailored to black-box deep natural-language models. Given a deep NLP model and the textual input data, T-EBAnO provides an objective, human-readable, domain-specific assessment of the reasons behind the automatic decision-making process. Specifically, the framework extracts sets of interpretable features mining the inner knowledge of the model. Then, it quantifies the influence of each feature during the prediction process by exploiting the novel normalized Perturbation Influence Relation index at the local level and the novel Global Absolute Influence and Global Relative Influence indexes at the global level. The effectiveness and the quality of the local and global explanations obtained with T-EBAnO are proved on (i) a sentiment analysis task performed by a fine-tuned BERT model, and (ii) a toxic comment classification task performed by an LSTM model.

Towards interpretability of Mixtures of Hidden Markov Models

Negar Safinianaini, Henrik Boström

Mixtures of Hidden Markov Models (MHMMs) are frequently used for clustering of sequential data. An important aspect of MHMMs, as of any clustering approach, is that they can be interpretable, allowing for novel insights to be gained from the data. However, without a proper way of measuring interpretability, the evaluation of novel contributions is difficult and it becomes practically impossible to devise techniques that directly optimize this property. In this work, an information-theoretic measure (entropy) is proposed for interpretability of MHMMs, and based on that, a novel approach to improve model interpretability is proposed, i.e., an entropy-regularized Expectation Maximization (EM) algorithm. The new approach aims for reducing the entropy of the Markov chains (involving state transition matrices) within an MHMM, i.e., assigning higher weights to common state transitions during clustering. It is argued that this entropy reduction, in general, leads to improved interpretability since the most influential and important state transitions of the clusters can be more easily identified. An empirical investigation shows that it is possible to improve the interpretability of MHMMs, as measured by entropy, without sacrificing (but rather improving) clustering performance and computational costs, as measured by the v-measure and number of EM iterations, respectively.

When human cognition is modeled in Philosophy and Cognitive Science, there is a pervasive idea that humans employ mental representations in order to navigate the world and make predictions about outcomes of future actions. By understanding how these representational structures work, we not only understand more about human cognition but also gain a better understanding for how humans rationalise and explain decisions. This has an influencing effect on explainable AI, where the goal is to provide explanations of computer decision-making for a human audience. We show that the Contextual Importance and Utility method for XAI share an overlap with the current new wave of action-oriented predictive representational structures, in ways that makes CIU a reliable tool for creating explanations that humans can relate to and trust.

Investigating ADR mechanisms with knowledge graph mining and explainable AI

Emmanuel Bresso, Pierre Monnin, Cédric Bousquet +5

Adverse Drug Reactions (ADRs) are characterized within randomized clinical trials and postmarketing pharmacovigilance, but their molecular mechanism remains unknown in most cases. Aside from clinical trials, many elements of knowledge about drug ingredients are available in open-access knowledge graphs. In addition, drug classifications that label drugs as either causative or not for several ADRs, have been established. We propose to mine knowledge graphs for identifying biomolecular features that may enable reproducing automatically expert classifications that distinguish drug causative or not for a given type of ADR. In an explainable AI perspective, we explore simple classification techniques such as Decision Trees and Classification Rules because they provide human-readable models, which explain the classification itself, but may also provide elements of explanation for molecular mechanisms behind ADRs. In summary, we mine a knowledge graph for features; we train classifiers at distinguishing, drugs associated or not with ADRs; we isolate features that are both efficient in reproducing expert classifications and interpretable by experts (i.e., Gene Ontology terms, drug targets, or pathway names); and we manually evaluate how they may be explanatory. Extracted features reproduce with a good fidelity classifications of drugs causative or not for DILI and SCAR. Experts fully agreed that 73% and 38% of the most discriminative features are possibly explanatory for DILI and SCAR, respectively; and partially agreed (2/3) for 90% and 77% of them. Knowledge graphs provide diverse features to enable simple and explainable models to distinguish between drugs that are causative or not for ADRs. In addition to explaining classifications, most discriminative features appear to be good candidates for investigating ADR mechanisms further.

In a variety of application settings, the user preference for a planning task - the precise optimization objective - is difficult to elicit. One possible remedy is planning as an iterative process, allowing the user to iteratively refine and modify example plans. A key step to support such a process are explanations, answering user questions about the current plan. In particular, a relevant kind of question is "Why does the plan you suggest not satisfy $p$?", where p is a plan property desirable to the user. Note that such a question pertains to plan space, i.e., the set of possible alternative plans. Adopting the recent approach to answer such questions in terms of plan-property dependencies, here we implement a tool and user interface for human-guided iterative planning including plan-space explanations. The tool runs in standard Web browsers, and provides simple user interfaces for both developers and users. We conduct a first user study, whose outcome indicates the usefulness of plan-property dependency explanations in iterative planning.

Provenance-Based Assessment of Plans in Context

Scott E. Friedman, Robert P. Goldman, Richard G. Freedman +3

Many real-world planning domains involve diverse information sources, external entities, and variable-reliability agents, all of which may impact the confidence, risk, and sensitivity of plans. Humans reviewing a plan may lack context about these factors; however, this information is available during the domain generation, which means it can also be interwoven into the planner and its resulting plans. This paper presents a provenance-based approach to explaining automated plans. Our approach (1) extends the SHOP3 HTN planner to generate dependency information, (2) transforms the dependency information into an established PROV-O representation, and (3) uses graph propagation and TMS-inspired algorithms to support dynamic and counter-factual assessment of information flow, confidence, and support. We qualified our approach's explanatory scope with respect to explanation targets from the automated planning literature and the information analysis literature, and we demonstrate its ability to assess a plan's pertinence, sensitivity, risk, assumption support, diversity, and relative confidence.

The work reported here addresses the capacity of psychophysiological sensors and measures using Electroencephalogram (EEG) and Galvanic Skin Response (GSR) to detect levels of trust for humans using AI-supported Human-Machine Interaction (HMI). Improvements to the analysis of EEG and GSR data may create models that perform as well, or better than, traditional tools. A challenge to analyzing the EEG and GSR data is the large amount of training data required due to a large number of variables in the measurements. Researchers have routinely used standard machine-learning classifiers like artificial neural networks (ANN), support vector machines (SVM), and K-nearest neighbors (KNN). Traditionally, these have provided few insights into which features of the EEG and GSR data facilitate the more and least accurate predictions - thus making it harder to improve the HMI and human-machine trust relationship. A key ingredient to applying trust-sensor research results to practical situations and monitoring trust in work environments is the understanding of which key features are contributing to trust and then reducing the amount of data needed for practical applications. We used the Local Interpretable Model-agnostic Explanations (LIME) model as a process to reduce the volume of data required to monitor and enhance trust in HMI systems - a technology that could be valuable for governmental and public sector applications. Explainable AI can make HMI systems transparent and promote trust. From customer service in government agencies and community-level non-profit public service organizations to national military and cybersecurity institutions, many public sector organizations are increasingly concerned to have effective and ethical HMI with services that are trustworthy, unbiased, and free of unintended negative consequences.

We examine the problem of explainable AI (xAI) and explore what delivering xAI means in practice, particularly in contexts that involve formal or informal and ad-hoc collaboration where agency and accountability in decision-making are achieved and sustained interactionally. We use an example from an earlier study of collaborative decision-making in screening mammography and the difficulties users faced when trying to interpret the behavior of an AI tool to illustrate the challenges of delivering usable and effective xAI. We conclude by setting out a study programme for future research to help advance our understanding of xAI requirements for safe and ethical AI.

The overarching goal of Explainable AI is to develop systems that not only exhibit intelligent behaviours, but also are able to explain their rationale and reveal insights. In explainable machine learning, methods that produce a high level of prediction accuracy as well as transparent explanations are valuable. In this work, we present an explainable classification method. Our method works by first constructing a symbolic Knowledge Base from the training data, and then performing probabilistic inferences on such Knowledge Base with linear programming. Our approach achieves a level of learning performance comparable to that of traditional classifiers such as random forests, support vector machines and neural networks. It identifies decisive features that are responsible for a classification as explanations and produces results similar to the ones found by SHAP, a state of the art Shapley Value based method. Our algorithms perform well on a range of synthetic and non-synthetic data sets.