The growing sophistication of Business Email Compromise (BEC) and spear phishing attacks poses significant challenges to organizations worldwide. The techniques featured in traditional spam and phishing detection are insufficient due to the tailored nature of modern BEC attacks as they often blend in with the regular benign traffic. Recent advances in machine learning, particularly in Natural Language Understanding (NLU), offer a promising avenue for combating such attacks but in a practical system, due to limitations such as data availability, operational costs, verdict explainability requirements or a need to robustly evolve the system, it is essential to combine multiple approaches together. We present CAPE, a comprehensive and efficient system for BEC detection that has been proven in a production environment for a period of over two years. Rather than being a single model, CAPE is a system that combines independent ML models and algorithms detecting BEC-related behaviors across various email modalities such as text, images, metadata and the email's communication context. This decomposition makes CAPE's verdicts naturally explainable. In the paper, we describe the design principles and constraints behind its architecture, as well as the challenges of model design, evaluation and adapting the system continuously through a Bayesian approach that combines limited data with domain knowledge. Furthermore, we elaborate on several specific behavioral detectors, such as those based on Transformer neural architectures.

As the complexity of learning tasks surges, modern machine learning encounters a new constrained learning paradigm characterized by more intricate and data-driven function constraints. Prominent applications include Neyman-Pearson classification (NPC) and fairness classification, which entail specific risk constraints that render standard projection-based training algorithms unsuitable. Gradient boosting machines (GBMs) are among the most popular algorithms for supervised learning; however, they are generally limited to unconstrained settings. In this paper, we adapt the GBM for constrained learning tasks within the framework of Bregman proximal algorithms. We introduce a new Bregman primal-dual method with a global optimality guarantee when the learning objective and constraint functions are convex. In cases of nonconvex functions, we demonstrate how our algorithm remains effective under a Bregman proximal point framework. Distinct from existing constrained learning algorithms, ours possess a unique advantage in their ability to seamlessly integrate with publicly available GBM implementations such as XGBoost (Chen and Guestrin, 2016) and LightGBM (Ke et al., 2017), exclusively relying on their public interfaces. We provide substantial experimental evidence to showcase the effectiveness of the Bregman algorithm framework. While our primary focus is on NPC and fairness ML, our framework holds significant potential for a broader range of constrained learning applications. The source code is currently freely available at https://github.com/zhenweilin/ConstrainedGBM}{https://github.com/zhenweilin/ConstrainedGBM.

Research has shown that language models (LMs) are socially biased. However, toxicity and offensive stereotyping bias in LMs are understudied. In this paper, we investigate the systematic offensive stereotype (SOS) bias in LMs. We propose a method to measure it. Then, we validate the SOS bias and investigate the effectiveness of debias methods from the literature on removing it. Finally, we investigate the impact of the SOS bias in LMs on their performance and their fairness on the task of hate speech detection. Our results suggest that all the inspected LMs are SOS biased. The results suggest that the SOS bias in LMs is reflective of the hate experienced online by the inspected marginalized groups. The results indicate that removing the SOS bias in LMs, using a popular debias method from the literature, leads to worse SOS bias scores. Finally, Our results show no strong evidence that the SOS bias in LMs is impactful on their performance on hate speech detection. On the other hand, there is evidence that the SOS bias in LMs is impactful on their fairness.

A plethora of outlier detectors have been explored in the time series domain, however, in a business sense, not all outliers are anomalies of interest. Existing anomaly detection solutions are confined to certain outlier detectors limiting their applicability to broader anomaly detection use cases. Network KPIs (Key Performance Indicators) tend to exhibit stochastic behaviour producing statistical outliers, most of which do not adversely affect business operations. Thus, a heuristic is required to capture the business definition of an anomaly for time series KPI. This article proposes an Adaptive Thresholding Heuristic (ATH) to dynamically adjust the detection threshold based on the local properties of the data distribution and adapt to changes in time series patterns. The heuristic derives the threshold based on the expected periodicity and the observed proportion of anomalies minimizing false positives and addressing concept drift. ATH can be used in conjunction with any underlying seasonality decomposition method and an outlier detector that yields an outlier score. This method has been tested on EON1-Cell-U, a labeled KPI anomaly dataset produced by Ericsson, to validate our hypothesis. Experimental results show that ATH is computationally efficient making it scalable for near real time anomaly detection and flexible with multiple forecasters and outlier detectors.

Multimodal learning often outperforms its unimodal counterparts by exploiting unimodal contributions and cross-modal interactions. However, focusing only on integrating multimodal features into a unified comprehensive representation overlooks the unimodal characteristics. In real data, the contributions of modalities can vary from instance to instance, and they often reinforce or conflict with each other. In this study, we introduce a novel \text{MultiModal} loss paradigm for multimodal learning, which subgroups instances according to their unimodal contributions. \text{MultiModal} loss can prevent inefficient learning caused by overfitting and efficiently optimize multimodal models. On synthetic data, \text{MultiModal} loss demonstrates improved classification performance by subgrouping difficult instances within certain modalities. On four real multimodal datasets, our loss is empirically shown to improve the performance of recent models. Ablation studies verify the effectiveness of our loss. Additionally, we show that our loss generates a reliable prediction score for each modality, which is essential for subgrouping. Our \text{MultiModal} loss is a novel loss function to subgroup instances according to the contribution of modalities in multimodal learning and is applicable to a variety of multimodal models with unimodal decisions. Our code is available at https://github.com/SehwanMoon/MultiModalLoss.

This paper introduces FairDP, a novel mechanism designed to achieve certified fairness with differential privacy (DP). FairDP independently trains models for distinct individual groups, using group-specific clipping terms to assess and bound the disparate impacts of DP. Throughout the training process, the mechanism progressively integrates knowledge from group models to formulate a comprehensive model that balances privacy, utility, and fairness in downstream tasks. Extensive theoretical and empirical analyses validate the efficacy of FairDP and improved trade-offs between model utility, privacy, and fairness compared with existing methods.

Out-of-distribution detection is one of the most critical issue in the deployment of machine learning. The data analyst must assure that data in operation should be compliant with the training phase as well as understand if the environment has changed in a way that autonomous decisions would not be safe anymore. The method of the paper is based on eXplainable Artificial Intelligence (XAI); it takes into account different metrics to identify any resemblance between in-distribution and out of, as seen by the XAI model. The approach is non-parametric and distributional assumption free. The validation over complex scenarios (predictive maintenance, vehicle platooning, covert channels in cybersecurity) corroborates both precision in detection and evaluation of training-operation conditions proximity. Results are available via open source and open data at the following link: https://github.com/giacomo97cnr/Rule-based-ODD.

Continual learning with an increasing number of classes is a challenging task. The difficulty rises when each example is presented exactly once, which requires the model to learn online. Recent methods with classic parameter optimization procedures have been shown to struggle in such setups or have limitations like non-differentiable components or memory buffers. For this reason, we present the fully differentiable ensemble method that allows us to efficiently train an ensemble of neural networks in the end-to-end regime. The proposed technique achieves SOTA results without a memory buffer and clearly outperforms the reference methods. The conducted experiments have also shown a significant increase in the performance for small ensembles, which demonstrates the capability of obtaining relatively high classification accuracy with a reduced number of classifiers.

Conditional selective inference requires an exact characterization of the selection event, which is often unavailable except for a few examples like the lasso. This work addresses this challenge by introducing a generic approach to estimate the selection event, facilitating feasible inference conditioned on the selection event. The method proceeds by repeatedly generating bootstrap data and running the selection algorithm on the new datasets. Using the outputs of the selection algorithm, we can estimate the selection probability as a function of certain summary statistics. This leads to an estimate of the distribution of the data conditioned on the selection event, which forms the basis for conditional selective inference. We provide a theoretical guarantee assuming both asymptotic normality of relevant statistics and accurate estimation of the selection probability. The applicability of the proposed method is demonstrated through a variety of problems that lack exact characterizations of selection, where conditional selective inference was previously infeasible.

Histopathological images are the gold standard for diagnosing liver cancer. However, the accuracy of fully digital diagnosis in computational pathology needs to be improved. In this paper, in order to solve the problem of multi-label and low classification accuracy of histopathology images, we propose a locally deep convolutional Swim framework (LDCSF) to classify multi-label histopathology images. In order to be able to provide local field of view diagnostic results, we propose the LDCSF model, which consists of a Swin transformer module, a local depth convolution (LDC) module, a feature reconstruction (FR) module, and a ResNet module. The Swin transformer module reduces the amount of computation generated by the attention mechanism by limiting the attention to each window. The LDC then reconstructs the attention map and performs convolution operations in multiple channels, passing the resulting feature map to the next layer. The FR module uses the corresponding weight coefficient vectors obtained from the channels to dot product with the original feature map vector matrix to generate representative feature maps. Finally, the residual network undertakes the final classification task. As a result, the classification accuracy of LDCSF for interstitial area, necrosis, non-tumor and tumor reached 0.9460, 0.9960, 0.9808, 0.9847, respectively. Finally, we use the results of multi-label pathological image classification to calculate the tumor-to-stromal ratio, which lays the foundation for the analysis of the microenvironment of liver cancer histopathological images. Second, we released a multilabel histopathology image of liver cancer, our code and data are available at https://github.com/panliangrui/LSF.