An ensuing challenge in Artificial Intelligence (AI) is the perceived difficulty in interpreting sophisticated machine learning models, whose ever-increasing complexity makes it hard for such models to be understood, trusted and thus accepted by human beings. The lack, if not complete absence, of interpretability for these so-called black-box models can lead to serious economic and ethical consequences, thereby hindering the development and deployment of AI in wider fields, particularly in those involving critical and regulatory applications. Yet, the building services industry is a highly-regulated domain requiring transparency and decision-making processes that can be understood and trusted by humans. To this end, the design and implementation of autonomous Heating, Ventilation and Air Conditioning systems for the automatic but concurrently interpretable optimisation of energy efficiency and room thermal comfort is of topical interest. This work therefore presents an interpretable machine learning model aimed at predicting room temperature in non-domestic buildings, for the purpose of optimising the use of the installed HVAC system. We demonstrate experimentally that the proposed model can accurately forecast room temperatures eight hours ahead in real-time by taking into account historical RT information, as well as additional environmental and time-series features. In this paper, an enhanced feature engineering process is conducted based on the Exploratory Data Analysis results. Furthermore, beyond the commonly used Interpretable Machine Learning techniques, we propose a Permutation Feature-based Frequency Response Analysis (PF-FRA) method for quantifying the contributions of the different predictors in the frequency domain. Based on the generated reason codes, we find that the historical RT feature is the dominant factor that has most impact on the model prediction.
Machine learning models can assist with metamaterials design by approximating computationally expensive simulators or solving inverse design problems. However, past work has usually relied on black box deep neural networks, whose reasoning processes are opaque and require enormous datasets that are expensive to obtain. In this work, we develop two novel machine learning approaches to metamaterials discovery that have neither of these disadvantages. These approaches, called shape-frequency features and unit-cell templates, can discover 2D metamaterials with user-specified frequency band gaps. Our approaches provide logical rule-based conditions on metamaterial unit-cells that allow for interpretable reasoning processes, and generalize well across design spaces of different resolutions. The templates also provide design flexibility where users can almost freely design the fine resolution features of a unit-cell without affecting the user's desired band gap.
Automated hyperparameter optimization (HPO) can support practitioners to obtain peak performance in machine learning models. However, there is often a lack of valuable insights into the effects of different hyperparameters on the final model performance. This lack of explainability makes it difficult to trust and understand the automated HPO process and its results. We suggest using interpretable machine learning (IML) to gain insights from the experimental data obtained during HPO with Bayesian optimization (BO). BO tends to focus on promising regions with potential high-performance configurations and thus induces a sampling bias. Hence, many IML techniques, such as the partial dependence plot (PDP), carry the risk of generating biased interpretations. By leveraging the posterior uncertainty of the BO surrogate model, we introduce a variant of the PDP with estimated confidence bands. We propose to partition the hyperparameter space to obtain more confident and reliable PDPs in relevant sub-regions. In an experimental study, we provide quantitative evidence for the increased quality of the PDPs within sub-regions.
Despite the widespread use of Knowledge Graph Embeddings (KGE), little is known about the security vulnerabilities that might disrupt their intended behaviour. We study data poisoning attacks against KGE models for link prediction. These attacks craft adversarial additions or deletions at training time to cause model failure at test time. To select adversarial deletions, we propose to use the model-agnostic instance attribution methods from Interpretable Machine Learning, which identify the training instances that are most influential to a neural model's predictions on test instances. We use these influential triples as adversarial deletions. We further propose a heuristic method to replace one of the two entities in each influential triple to generate adversarial additions. Our experiments show that the proposed strategies outperform the state-of-art data poisoning attacks on KGE models and improve the MRR degradation due to the attacks by up to 62% over the baselines.
Post-hoc global/local feature attribution methods are progressively being employed to understand the decisions of complex machine learning models. Yet, because of limited amounts of data, it is possible to obtain a diversity of models with good empirical performance but that provide very different explanations for the same prediction, making it hard to derive insight from them. In this work, instead of aiming at reducing the under-specification of model explanations, we fully embrace it and extract logical statements about feature attributions that are consistent across all models with good empirical performance (i.e. all models in the Rashomon Set). We show that partial orders of local/global feature importance arise from this methodology enabling more nuanced interpretations by allowing pairs of features to be incomparable when there is no consensus on their relative importance. We prove that every relation among features present in these partial orders also holds in the rankings provided by existing approaches. Finally, we present three use cases employing hypothesis spaces with tractable Rashomon Sets (Additive models, Kernel Ridge, and Random Forests) and show that partial orders allow one to extract consistent local and global interpretations of models despite their under-specification.
Interpretable machine learning has been focusing on explaining final models that optimize performance. The current state-of-the-art is the Shapley additive explanations (SHAP) that locally explains variable impact on individual predictions, and it is recently extended for a global assessment across the dataset. Recently, Dong and Rudin proposed to extend the investigation to models from the same class as the final model that are "good enough", and identified a previous overclaim of variable importance based on a single model. However, this method does not directly integrate with existing Shapley-based interpretations. We close this gap by proposing a Shapley variable importance cloud that pools information across good models to avoid biased assessments in SHAP analyses of final models, and communicate the findings via novel visualizations. We demonstrate the additional insights gain compared to conventional explanations and Dong and Rudin's method using criminal justice and electronic medical records data.
Disinformation through fake news is an ongoing problem in our society and has become easily spread through social media. The most cost and time effective way to filter these large amounts of data is to use a combination of human and technical interventions to identify it. From a technical perspective, Natural Language Processing (NLP) is widely used in detecting fake news. Social media companies use NLP techniques to identify the fake news and warn their users, but fake news may still slip through undetected. It is especially a problem in more localised contexts (outside the United States of America). How do we adjust fake news detection systems to work better for local contexts such as in South Africa. In this work we investigate fake news detection on South African websites. We curate a dataset of South African fake news and then train detection models. We contrast this with using widely available fake news datasets (from mostly USA website). We also explore making the datasets more diverse by combining them and observe the differences in behaviour in writing between nations' fake news using interpretable machine learning.
Autonomous vehicles must balance a complex set of objectives. There is no consensus on how they should do so, nor on a model for specifying a desired driving behavior. We created a dataset to help address some of these questions in a limited operating domain. The data consists of 92 traffic scenarios, with multiple ways of traversing each scenario. Multiple annotators expressed their preference between pairs of scenario traversals. We used the data to compare an instance of a rulebook, carefully hand-crafted independently of the dataset, with several interpretable machine learning models such as Bayesian networks, decision trees, and logistic regression trained on the dataset. To compare driving behavior, these models use scores indicating by how much different scenario traversals violate each of 14 driving rules. The rules are interpretable and designed by subject-matter experts. First, we found that these rules were enough for these models to achieve a high classification accuracy on the dataset. Second, we found that the rulebook provides high interpretability without excessively sacrificing performance. Third, the data pointed to possible improvements in the rulebook and the rules, and to potential new rules. Fourth, we explored the interpretability vs performance trade-off by also training non-interpretable models such as a random forest. Finally, we make the dataset publicly available to encourage a discussion from the wider community on behavior specification for AVs. Please find it at github.com/bassam-motional/Reasonable-Crowd.
Background: Medical decision-making impacts both individual and public health. Clinical scores are commonly used among a wide variety of decision-making models for determining the degree of disease deterioration at the bedside. AutoScore was proposed as a useful clinical score generator based on machine learning and a generalized linear model. Its current framework, however, still leaves room for improvement when addressing unbalanced data of rare events. Methods: Using machine intelligence approaches, we developed AutoScore-Imbalance, which comprises three components: training dataset optimization, sample weight optimization, and adjusted AutoScore. All scoring models were evaluated on the basis of their area under the curve (AUC) in the receiver operating characteristic analysis and balanced accuracy (i.e., mean value of sensitivity and specificity). By utilizing a publicly accessible dataset from Beth Israel Deaconess Medical Center, we assessed the proposed model and baseline approaches in the prediction of inpatient mortality. Results: AutoScore-Imbalance outperformed baselines in terms of AUC and balanced accuracy. The nine-variable AutoScore-Imbalance sub-model achieved the highest AUC of 0.786 (0.732-0.839) while the eleven-variable original AutoScore obtained an AUC of 0.723 (0.663-0.783), and the logistic regression with 21 variables obtained an AUC of 0.743 (0.685-0.800). The AutoScore-Imbalance sub-model (using down-sampling algorithm) yielded an AUC of 0. 0.771 (0.718-0.823) with only five variables, demonstrating a good balance between performance and variable sparsity. Conclusions: The AutoScore-Imbalance tool has the potential to be applied to highly unbalanced datasets to gain further insight into rare medical events and to facilitate real-world clinical decision-making.
Though deep learning has shown successful performance in classifying the label and severity stage of certain diseases, most of them give few explanations on how to make predictions. Inspired by Koch's Postulates, the foundation in evidence-based medicine (EBM) to identify the pathogen, we propose to exploit the interpretability of deep learning application in medical diagnosis. By determining and isolating the neuron activation patterns on which diabetic retinopathy (DR) detector relies to make decisions, we demonstrate the direct relation between the isolated neuron activation and lesions for a pathological explanation. To be specific, we first define novel pathological descriptors using activated neurons of the DR detector to encode both spatial and appearance information of lesions. Then, to visualize the symptom encoded in the descriptor, we propose Patho-GAN, a new network to synthesize medically plausible retinal images. By manipulating these descriptors, we could even arbitrarily control the position, quantity, and categories of generated lesions. We also show that our synthesized images carry the symptoms directly related to diabetic retinopathy diagnosis. Our generated images are both qualitatively and quantitatively superior to the ones by previous methods. Besides, compared to existing methods that take hours to generate an image, our second level speed endows the potential to be an effective solution for data augmentation.
Gradual argumentation frameworks represent arguments and their relationships in a weighted graph. Their graphical structure and intuitive semantics makes them a potentially interesting tool for interpretable machine learning. It has been noted recently that their mechanics are closely related to neural networks, which allows learning their weights from data by standard deep learning frameworks. As a first proof of concept, we propose a genetic algorithm to simultaneously learn the structure of argumentative classification models. To obtain a well interpretable model, the fitness function balances sparseness and accuracy of the classifier. We discuss our algorithm and present first experimental results on standard benchmarks from the UCI machine learning repository. Our prototype learns argumentative classification models that are comparable to decision trees in terms of learning performance and interpretability.
Although modern machine learning and deep learning methods allow for complex and in-depth data analytics, the predictive models generated by these methods are often highly complex, and lack transparency. Explainable AI (XAI) methods are used to improve the interpretability of these complex models, and in doing so improve transparency. However, the inherent fitness of these explainable methods can be hard to evaluate. In particular, methods to evaluate the fidelity of the explanation to the underlying black box require further development, especially for tabular data. In this paper, we (a) propose a three phase approach to developing an evaluation method; (b) adapt an existing evaluation method primarily for image and text data to evaluate models trained on tabular data; and (c) evaluate two popular explainable methods using this evaluation method. Our evaluations suggest that the internal mechanism of the underlying predictive model, the internal mechanism of the explainable method used and model and data complexity all affect explanation fidelity. Given that explanation fidelity is so sensitive to context and tools and data used, we could not clearly identify any specific explainable method as being superior to another.
Scoring systems are highly interpretable and widely used to evaluate time-to-event outcomes in healthcare research. However, existing time-to-event scores are predominantly created ad-hoc using a few manually selected variables based on clinician's knowledge, suggesting an unmet need for a robust and efficient generic score-generating method. AutoScore was previously developed as an interpretable machine learning score generator, integrated both machine learning and point-based scores in the strong discriminability and accessibility. We have further extended it to time-to-event data and developed AutoScore-Survival, for automatically generating time-to-event scores with right-censored survival data. Random survival forest provides an efficient solution for selecting variables, and Cox regression was used for score weighting. We illustrated our method in a real-life study of 90-day mortality of patients in intensive care units and compared its performance with survival models (i.e., Cox) and the random survival forest. The AutoScore-Survival-derived scoring model was more parsimonious than survival models built using traditional variable selection methods (e.g., penalized likelihood approach and stepwise variable selection), and its performance was comparable to survival models using the same set of variables. Although AutoScore-Survival achieved a comparable integrated area under the curve of 0.782 (95% CI: 0.767-0.794), the integer-valued time-to-event scores generated are favorable in clinical applications because they are easier to compute and interpret. Our proposed AutoScore-Survival provides an automated, robust and easy-to-use machine learning-based clinical score generator to studies of time-to-event outcomes. It provides a systematic guideline to facilitate the future development of time-to-event scores for clinical applications.
The field of Explainable Artificial Intelligence (XAI) aims to build explainable and interpretable machine learning (or deep learning) methods without sacrificing prediction performance. Convolutional Neural Networks (CNNs) have been successful in making predictions, especially in image classification. However, these famous deep learning models use tens of millions of parameters based on a large number of pre-trained filters which have been repurposed from previous data sets. We propose a novel Interaction-based Convolutional Neural Network (ICNN) that does not make assumptions about the relevance of local information. Instead, we use a model-free Influence Score (I-score) to directly extract the influential information from images to form important variable modules. We demonstrate that the proposed method produces state-of-the-art prediction performance of 99.8% on a real-world data set classifying COVID-19 Chest X-ray images without sacrificing the explanatory power of the model. This proposed design can efficiently screen COVID-19 patients before human diagnosis, and will be the benchmark for addressing future XAI problems in large-scale data sets.
Counterfactual explanations are usually generated through heuristics that are sensitive to the search's initial conditions. The absence of guarantees of performance and robustness hinders trustworthiness. In this paper, we take a disciplined approach towards counterfactual explanations for tree ensembles. We advocate for a model-based search aiming at "optimal" explanations and propose efficient mixed-integer programming approaches. We show that isolation forests can be modeled within our framework to focus the search on plausible explanations with a low outlier score. We provide comprehensive coverage of additional constraints that model important objectives, heterogeneous data types, structural constraints on the feature space, along with resource and actionability restrictions. Our experimental analyses demonstrate that the proposed search approach requires a computational effort that is orders of magnitude smaller than previous mathematical programming algorithms. It scales up to large data sets and tree ensembles, where it provides, within seconds, systematic explanations grounded on well-defined models solved to optimality.
In this article, we introduce a novel variant of the Tsetlin machine (TM) that randomly drops clauses, the key learning elements of a TM. In effect, TM with drop clause ignores a random selection of the clauses in each epoch, selected according to a predefined probability. In this way, additional stochasticity is introduced in the learning phase of TM. To explore the effects drop clause has on accuracy, training time, interpretability and robustness, we conduct extensive experiments on nine benchmark datasets in natural language processing~(NLP) (IMDb, R8, R52, MR and TREC) and image classification (MNIST, Fashion MNIST, CIFAR-10 and CIFAR-100). Our proposed model outperforms baseline machine learning algorithms by a wide margin and achieves competitive performance in comparison with recent deep learning model such as BERT and AlexNET-DFA. In brief, we observe up to +10% increase in accuracy and 2x to 4x faster learning compared with standard TM. We further employ the Convolutional TM to document interpretable results on the CIFAR datasets, visualizing how the heatmaps produced by the TM become more interpretable with drop clause. We also evaluate how drop clause affects learning robustness by introducing corruptions and alterations in the image/language test data. Our results show that drop clause makes TM more robust towards such changes.
Explaining deep learning model inferences is a promising venue for scientific understanding, improving safety, uncovering hidden biases, evaluating fairness, and beyond, as argued by many scholars. One of the principal benefits of counterfactual explanations is allowing users to explore "what-if" scenarios through what does not and cannot exist in the data, a quality that many other forms of explanation such as heatmaps and influence functions are inherently incapable of doing. However, most previous work on generative explainability cannot disentangle important concepts effectively, produces unrealistic examples, or fails to retain relevant information. We propose a novel approach, DISSECT, that jointly trains a generator, a discriminator, and a concept disentangler to overcome such challenges using little supervision. DISSECT generates Concept Traversals (CTs), defined as a sequence of generated examples with increasing degrees of concepts that influence a classifier's decision. By training a generative model from a classifier's signal, DISSECT offers a way to discover a classifier's inherent "notion" of distinct concepts automatically rather than rely on user-predefined concepts. We show that DISSECT produces CTs that (1) disentangle several concepts, (2) are influential to a classifier's decision and are coupled to its reasoning due to joint training (3), are realistic, (4) preserve relevant information, and (5) are stable across similar inputs. We validate DISSECT on several challenging synthetic and realistic datasets where previous methods fall short of satisfying desirable criteria for interpretability and show that it performs consistently well and better than existing methods. Finally, we present experiments showing applications of DISSECT for detecting potential biases of a classifier and identifying spurious artifacts that impact predictions.
Explaining the decisions of machine learning models is becoming a necessity in many areas where trust in ML models decision is key to their accreditation/adoption. The ability to explain models decisions also allows to provide diagnosis in addition to the model decision, which is highly valuable in scenarios such as fault detection. Unfortunately, high-performance models do not exhibit the necessary transparency to make their decisions fully understandable. And the black-boxes approaches, which are used to explain such model decisions, suffer from a lack of accuracy in tracing back the exact cause of a model decision regarding a given input. Indeed, they do not have the ability to explicitly describe the decision regions of the model around that input, which is necessary to determine what influences the model towards one decision or the other. We thus asked ourselves the question: is there a category of high-performance models among the ones currently used for which we could explicitly and exactly characterise the decision regions in the input feature space using a geometrical characterisation? Surprisingly we came out with a positive answer for any model that enters the category of tree ensemble models, which encompasses a wide range of high-performance models such as XGBoost, LightGBM, random forests ... We could derive an exact geometrical characterisation of their decision regions under the form of a collection of multidimensional intervals. This characterisation makes it straightforward to compute the optimal counterfactual (CF) example associated with a query point. We demonstrate several possibilities of the approach, such as computing the CF example based only on a subset of features. This allows to obtain more plausible explanations by adding prior knowledge about which variables the user can control. An adaptation to CF reasoning on regression problems is also envisaged.
We have built a computational model for individual aging trajectories of health and survival, which contains physical, functional, and biological variables, and is conditioned on demographic, lifestyle, and medical background information. We combine techniques of modern machine learning with an interpretable interaction network, where health variables are coupled by explicit pair-wise interactions within a stochastic dynamical system. Our dynamic joint interpretable network (DJIN) model is scalable to large longitudinal data sets, is predictive of individual high-dimensional health trajectories and survival from baseline health states, and infers an interpretable network of directed interactions between the health variables. The network identifies plausible physiological connections between health variables as well as clusters of strongly connected health variables. We use English Longitudinal Study of Aging (ELSA) data to train our model and show that it performs better than multiple dedicated linear models for health outcomes and survival. We compare our model with flexible lower-dimensional latent-space models to explore the dimensionality required to accurately model aging health outcomes. Our DJIN model can be used to generate synthetic individuals that age realistically, to impute missing data, and to simulate future aging outcomes given arbitrary initial health states.
A growing number of approaches exist to generate explanations for image classification. However, few of these approaches are subjected to human-subject evaluations, partly because it is challenging to design controlled experiments with natural image datasets, as they leave essential factors out of the researcher's control. With our approach, researchers can describe their desired dataset with only a few parameters. Based on these, our library generates synthetic image data of two 3D abstract animals. The resulting data is suitable for algorithmic as well as human-subject evaluations. Our user study results demonstrate that our method can create biases predictive enough for a classifier and subtle enough to be noticeable only to every second participant inspecting the data visually. Our approach significantly lowers the barrier for conducting human subject evaluations, thereby facilitating more rigorous investigations into interpretable machine learning. For our library and datasets see, https://github.com/mschuessler/two4two/