Deep models have advanced prediction in many domains, but their lack of interpretability  remains a key barrier to the adoption in many real world applications. There exists a large  body of work aiming to help humans understand these black box functions to varying levels  of granularity – for example, through distillation, gradients, or adversarial examples. These  methods however, all tackle interpretability as a separate process after training. In this  work, we take a different approach and explicitly regularize deep models so that they are  well-approximated by processes that humans can step through in little time. Specifically,  we train several families of deep neural networks to resemble compact, axis-aligned decision  trees without significant compromises in accuracy. The resulting axis-aligned decision  functions uniquely make tree regularized models easy for humans to interpret. Moreover,  for situations in which a single, global tree is a poor estimator, we introduce a regional tree regularizer that encourages the deep model to resemble a compact, axis-aligned decision  tree in predefined, human-interpretable contexts. Using intuitive toy examples, benchmark  image datasets, and medical tasks for patients in critical care and with HIV, we demonstrate  that this new family of tree regularizers yield models that are easier for humans to simulate  than L1 or L2 penalties without sacrificing predictive power. 

COVID-19 frequently provokes pneumonia, which can be diagnosed using imaging exams. Chest X-ray (CXR) is often useful because it is cheap, fast, widespread, and uses less radiation. Here, we demonstrate the impact of lung segmentation in COVID-19 identification using CXR images and evaluate which contents of the image influenced the most. Semantic segmentation was performed using a U-Net CNN architecture, and the classification using three CNN architectures (VGG, ResNet, and Inception). Explainable Artificial Intelligence techniques were employed to estimate the impact of segmentation. A three-classes database was composed: lung opacity (pneumonia), COVID-19, and normal. We assessed the impact of creating a CXR image database from different sources, and the COVID-19 generalization from one source to another. The segmentation achieved a Jaccard distance of 0.034 and a Dice coefficient of 0.982. The classification using segmented images achieved an F1-Score of 0.88 for the multi-class setup, and 0.83 for COVID-19 identification. In the cross-dataset scenario, we obtained an F1-Score of 0.74 and an area under the ROC curve of 0.9 for COVID-19 identification using segmented images. Experiments support the conclusion that even after segmentation, there is a strong bias introduced by underlying factors from different sources.

In predictive tasks, real-world datasets often present different degrees of imbalanced (i.e., long-tailed or skewed) distributions. While the majority (the head) classes have sufficient samples, the minority (the tail) classes can be under-represented by a rather limited number of samples. Data pre-processing has been shown to be very effective in dealing with such problems. On one hand, data re-sampling is a common approach to tackling class imbalance. On the other hand, dimension reduction, which reduces the feature space, is a conventional technique for reducing noise and inconsistencies in a dataset. However, the possible synergy between feature selection and data re-sampling for high-performance imbalance classification has rarely been investigated before. To address this issue, we carry out a comprehensive empirical study on the joint influence of feature selection and re-sampling on two-class imbalance classification. Specifically, we study the performance of two opposite pipelines for imbalance classification by applying feature selection before or after data re-sampling. We conduct a large number of experiments, with a total of 9225 tests, on 52 publicly available datasets, using 9 feature selection methods, 6 re-sampling approaches for class imbalance learning, and 3 well-known classification algorithms. Experimental results show that there is no constant winner between the two pipelines; thus both of them should be considered to derive the best performing model for imbalance classification. We find that the performance of an imbalance classification model not only depends on the classifier adopted and the ratio between the number of majority and minority samples, but also depends on the ratio between the number of samples and features. Overall, this study should provide new reference value for researchers and practitioners in imbalance learning.

An interesting observation in artificial neural networks is their favorable generalization error despite typically being extremely overparameterized. It is well known that classical statistical learning methods often result in vacuous generalization errors in the case of overparameterized neural networks. Adopting the recently developed Neural Tangent (NT) kernel theory, we prove uniform generalization bounds for overparameterized neural networks in kernel regimes, when the true data generating model belongs to the reproducing kernel Hilbert space (RKHS) corresponding to the NT kernel. Importantly, our bounds capture the exact error rates depending on the differentiability of the activation functions. In order to establish these bounds, we propose the information gain of the NT kernel as a measure of complexity of the learning problem. Our analysis uses a Mercer decomposition of the NT kernel in the basis of spherical harmonics and the decay rate of the corresponding eigenvalues. As a byproduct of our results, we show the equivalence between the RKHS corresponding to the NT kernel and its counterpart corresponding to the Mat\'ern family of kernels, that induces a very general class of models. We further discuss the implications of our analysis for some recent results on the regret bounds for reinforcement learning algorithms, which use overparameterized neural networks.

This paper addresses the deep face recognition problem under an open-set protocol, where ideal face features are expected to have smaller maximal intra-class distance than minimal inter-class distance under a suitably chosen metric space. To this end, hyperspherical face recognition, as a promising line of research, has attracted increasing attention and gradually become a major focus in face recognition research. As one of the earliest works in hyperspherical face recognition, SphereFace explicitly proposed to learn face embeddings with large inter-class angular margin. However, SphereFace still suffers from severe training instability which limits its application in practice. In order to address this problem, we introduce a unified framework to understand large angular margin in hyperspherical face recognition. Under this framework, we extend the study of SphereFace and propose an improved variant with substantially better training stability -- SphereFace-R. Specifically, we propose two novel ways to implement the multiplicative margin, and study SphereFace-R under three different feature normalization schemes (no feature normalization, hard feature normalization and soft feature normalization). We also propose an implementation strategy -- "characteristic gradient detachment" -- to stabilize training. Extensive experiments on SphereFace-R show that it is consistently better than or competitive with state-of-the-art methods.

Gradient Boosting Machines (GBM) are among the go-to algorithms on tabular data, which produce state of the art results in many prediction tasks. Despite its popularity, the GBM framework suffers from a fundamental flaw in its base learners. Specifically, most implementations utilize decision trees that are typically biased towards categorical variables with large cardinalities. The effect of this bias was extensively studied over the years, mostly in terms of predictive performance. In this work, we extend the scope and study the effect of biased base learners on GBM feature importance (FI) measures. We show that although these implementation demonstrate highly competitive predictive performance, they still, surprisingly, suffer from bias in FI. By utilizing cross-validated (CV) unbiased base learners, we fix this flaw at a relatively low computational cost. We demonstrate the suggested framework in a variety of synthetic and real-world setups, showing a significant improvement in all GBM FI measures while maintaining relatively the same level of prediction accuracy.

If you report your classifier's performance as having Accuracy 94.8% and F1 92.3% on a test set, this doesn't mean much without knowing something about the size and composition of the test set. The margin of error of those performance measurements will vary a lot depending on the size of the test set, or, for an imbalanced dataset, primarily depending on how many independent instances of the minority class it contains (more copies of the same instances from oversampling doesn't help for this purpose). If you were able to collect another, independent test set of similar origin, the Accuracy and F1 of your model on this dataset are unlikely to be the same, but how much different might they plausibly be? A question similar to this is answered in statistics as the confidence interval of the measurement. If you just want to see how to invoke the code we provide to calculate confidence intervals, skip to the section "Calculate the results!" down below.

In Federated Learning (FL), a group of workers participate to build a global model under the coordination of one node, the chief. Regarding the cybersecurity of FL, some attacks aim at injecting the fabricated local model updates into the system. Some defenses are based on malicious worker detection and behavioral pattern analysis. In this context, without timely and dynamic monitoring methods, the chief cannot detect and remove the malicious or unreliable workers from the system. Our work emphasize the urgency to prepare the federated learning process for monitoring and eventually behavioral pattern analysis. We study the information inside the learning process in the early stages of training, propose a monitoring process and evaluate the monitoring period required. The aim is to analyse at what time is it appropriate to start the detection algorithm in order to remove the malicious or unreliable workers from the system and optimise the defense mechanism deployment. We tested our strategy on a behavioral pattern analysis defense applied to the FL process of different benchmark systems for text and image classification. Our results show that the monitoring process lowers false positives and false negatives and consequently increases system efficiency by enabling the distributed learning system to achieve better performance in the early stage of training.

In recent years, proposed studies on time-series anomaly detection (TAD) report high F1 scores on benchmark TAD datasets, giving the impression of clear improvements. However, most studies apply a peculiar evaluation protocol called point adjustment (PA) before scoring. In this paper, we theoretically and experimentally reveal that the PA protocol has a great possibility of overestimating the detection performance; that is, even a random anomaly score can easily turn into a state-of-the-art TAD method. Therefore, the comparison of TAD methods with F1 scores after the PA protocol can lead to misguided rankings. Furthermore, we question the potential of existing TAD methods by showing that an untrained model obtains comparable detection performance to the existing methods even without PA. Based on our findings, we propose a new baseline and an evaluation protocol. We expect that our study will help a rigorous evaluation of TAD and lead to further improvement in future researches.

In recent years, most fairness strategies in machine learning models focus on mitigating unwanted biases by assuming that the sensitive information is observed. However this is not always possible in practice. Due to privacy purposes and var-ious regulations such as RGPD in EU, many personal sensitive attributes are frequently not collected. We notice a lack of approaches for mitigating bias in such difficult settings, in particular for achieving classical fairness objectives such as Demographic Parity and Equalized Odds. By leveraging recent developments for approximate inference, we propose an approach to fill this gap. Based on a causal graph, we rely on a new variational auto-encoding based framework named SRCVAE to infer a sensitive information proxy, that serve for bias mitigation in an adversarial fairness approach. We empirically demonstrate significant improvements over existing works in the field. We observe that the generated proxy's latent space recovers sensitive information and that our approach achieves a higher accuracy while obtaining the same level of fairness on two real datasets, as measured using com-mon fairness definitions.