Deep learning-based anomaly detection (DAD) has been a hot topic of research in various domains. Despite being the most common data type, DAD for tabular data remains under-explored. Due to the scarcity of anomalies in real-world scenarios, deep semi-supervised learning methods have come to dominate, which build deep learning models and leverage a limited number of labeled anomalies and large-scale unlabeled data to improve their detection capabilities. However, existing works share two drawbacks. (1) Most of them simply treat the unlabeled samples as normal ones, ignoring the problem of label contamination, which is very common in real-world datasets. (2) Only very few works have designed models specifically for tabular data instead of migrating models from other domains to tabular data. Both of them will limit the model’s performance. In this work, we propose a feature interaction-based reinforcement learning for tabular anomaly detection, FIRTAD. FIRTAD incorporates a feature interaction module into a deep reinforcement learning framework; the former can model tabular data by learning a relationship among features, while the latter can effectively exploit available information and fully explore suspicious anomalies from the unlabeled samples. Extensive experiments on three datasets not only demonstrate its superiority over the state-of-art methods but also confirm its robustness to anomaly rarity, label contamination and unknown anomalies.

CTR (Click-Through Rate) prediction has attracted more and more attention from academia and industry for its significant contribution to revenue. In the last decade, learning feature interactions have become a mainstream research direction, and dozens of feature interaction-based models have been proposed for the CTR prediction task. The most common approach for existing models is to enumerate all possible feature interactions or to learn higher-order feature interactions by designing complex models. However, a simple enumeration will introduce meaningless and harmful interactions, and a complex model structure will bring a higher complexity. In this work, we propose a lightweight, yet effective model called the Gated Adaptive feature Interaction Network (GAIN). We devise a novel cross module to drop meaningless feature interactions and preserve informative ones. Our cross module consists of multiple gated units, each of which can independently learn an arbitrary-order feature interaction. We combine the cross module with a deep module into GAIN and conduct comparative experiments with state-of-the-art models on two public datasets to verify its validity. Our experimental results show that GAIN can achieve a comparable or even better performance compared to its competitors. Furthermore, in order to verify the effectiveness of the feature interactions learned by GAIN, we transfer learned interactions to other models, such as Logistic Regression (LR) and Factorization Machines (FM), and find out that their performance can be significantly improved.