Video anomaly detection (VAD) has long been studied as a crucial problem in public security and crime prevention. In recent years, weakly-supervised VAD (WVAD) has attracted considerable attention due to their easy annotation process and promising research results. While existing WVAD methods mainly tackle static datasets, the possibility that the domain of data can vary has been neglected. To adapt such domain shift, the continual learning (CL) perspective is required because otherwise additional training using only newly incoming data could easily cause performance degradation for previous data, i.e., forgetting. Therefore, we propose a brand-new approach, called Continual Anomaly Detection with Ensembles (CADE) that is the first work combining CL and WVAD viewpoints. Specifically, CADE uses the Dual-Generator (DG) to address data imbalance and label uncertainty in WVAD. We also found that forgetting exacerbates the "incompleteness" where the model becomes biased towards certain anomaly modes, leading to missed detections of various anomalies. To address this, we propose a Multi-Discriminator (MD) ensemble that captures missed anomalies in past scenes due to forgetting, using multiple models. Extensive experiments show that CADE significantly outperforms existing VAD methods on the common multi-scene VAD datasets, such as the ShanghaiTech, UCF-Crime, and Charlotte Anomaly Dataset.

Building high-quality recommender systems is challenging for new services and small companies, because of their sparse interactions. Cross-domain recommendations (CDRs) alleviate this issue by transferring knowledge from data in external domains. However, most existing CDRs leverage data from only a single external domain and serve only two domains. CDRs serving multiple domains require domain-shared entities (i.e., users and items) to transfer knowledge, which significantly limits their applications due to the hardness and privacy concerns of finding such entities. We therefore focus on a more general scenario, non-overlapped multi-target CDRs (NO-MTCDRs), which require no domain-shared entities and serve multiple domains. Existing methods require domain-shared users to learn user preferences and cannot work on NO-MTCDRs. We hence propose MITrans, a novel mutual information-based (MI-based) preference disentangling and transferring framework to improve recommendations for all domains. MITrans effectively leverages knowledge from multiple domains as well as learning both domain-shared and domain-specific preferences without using domain-shared users. In MITrans, we devise two novel MI constraints to disentangle domain-shared and domain-specific preferences. Moreover, we introduce a module that fuses domain-shared preferences in different domains and combines them with domain-specific preferences to improve recommendations. Our experimental results on two real-world datasets demonstrate the superiority of MITrans in terms of recommendation quality and application range against state-of-the-art overlapped and non-overlapped CDRs.

Existing research on task incremental learning in continual learning has primarily focused on preventing catastrophic forgetting (CF). Although several techniques have achieved learning with no CF, they attain it by letting each task monopolize a sub-network in a shared network, which seriously limits knowledge transfer (KT) and causes over-consumption of the network capacity, i.e., as more tasks are learned, the performance deteriorates. The goal of this paper is threefold: (1) overcoming CF, (2) encouraging KT, and (3) tackling the capacity problem. A novel technique (called SPG) is proposed that soft-masks (partially blocks) parameter updating in training based on the importance of each parameter to old tasks. Each task still uses the full network, i.e., no monopoly of any part of the network by any task, which enables maximum KT and reduction in capacity usage. To our knowledge, this is the first work that soft-masks a model at the parameter-level for continual learning. Extensive experiments demonstrate the effectiveness of SPG in achieving all three objectives. More notably, it attains significant transfer of knowledge not only among similar tasks (with shared knowledge) but also among dissimilar tasks (with little shared knowledge) while mitigating CF.

Deep learning-based recommender systems may lead to over-fitting when lacking training interaction data. This over-fitting significantly degrades recommendation performances. To address this data sparsity problem, cross-domain recommender systems (CDRSs) exploit the data from an auxiliary source domain to facilitate the recommendation on the sparse target domain. Most existing CDRSs rely on overlapping users or items to connect domains and transfer knowledge. However, matching users is an arduous task and may involve privacy issues when data comes from different companies, resulting in a limited application for the above CDRSs. Some studies develop CDRSs that require no overlapping users and items by transferring learned user interaction patterns. However, they ignore the bias in user interaction patterns between domains and hence suffer from an inferior performance compared with single-domain recommender systems. In this paper, based on the above findings, we propose a novel CDRS, namely semantic clustering enhanced debiasing graph neural recommender system (SCDGN), that requires no overlapping users and items and can handle the domain bias. More precisely, SCDGN semantically clusters items from both domains and constructs a cross-domain bipartite graph generated from item clusters and users. Then, the knowledge is transferred via this cross-domain user-cluster graph from source to the target. Furthermore, we design a debiasing graph convolutional layer for SCDGN to extract unbiased structural knowledge from the cross-domain user-cluster graph. Our Experimental results on three public datasets and a pair of proprietary datasets verify the effectiveness of SCDGN over stateof-the-art models in terms of cross-domain recommendations.

Knowledge graphs (KGs) are useful information sources to make machine learning efficient with human knowledge. Since KGs are often incomplete, KG completion has become an important problem to complete missing facts in KGs. Whereas most of the KG completion methods are conducted on a single KG, multiple KGs can be effective to enrich embedding space for KG completion. However, most of the recent studies have concentrated on entity alignment prediction and ignored KG-invariant semantics in multiple KGs that can improve the completion performance. In this paper, we propose a new multiple KG embedding method composed of intra-KG and inter-KG regularization to introduce KG-invariant semantics into KG embedding space using aligned entities between related KGs. The intra-KG regularization adjusts local distance between aligned and not-aligned entities using contrastive loss, while the inter-KG regularization globally correlates aligned entity embeddings between KGs using multi-view loss. Our experimental results demonstrate that our proposed method combining both regularization terms largely outperforms existing baselines in the KG completion task.

Recommender systems (RSs) have been extensively studied in academia and industry, while few works focus on flash sale recommendations. In flash sale scenarios, period-specific high discounts are applied on ordinal sales during each flash sale period to attract users. According to periodic sales strategies, available and discounted items change significantly across periods. Users are attracted by the high discounts and show a period-specific preference. Besides, the frequently changed available items provoke a cold-start problem, i.e., an impaired recommendation caused by lacking interactions. However, most existing RSs either cannot handle users’ period-specific preferences or suffer from the cold-start problem. Therefore, this work proposes a novel meta-learning-based RS to mitigate the cold-start problem and simultaneously handle users’ period-specific preferences. Moreover, we introduce a novel hierarchical meta-training algorithm to guide the learning of our recommendation model via period-and user-specific gradients. In this way, the learned model contains user- and period-shared knowledge and can fast adapt to the recommendations for new flash sale periods and users. To evaluate the effectiveness of our system in flash sale recommendations and non-flash sale recommendations, we conduct experiments on a real-world flash sale e-commerce dataset and a widely used recommendation dataset, considering both warm and cold scenarios. The experimental results show that our proposed model is remarkably improved over the current state-of-the-art methods in flash sale recommendations and most of the non-flash sale cold-start recommendations.

With the deployment of smart homes, we find that human activity recognition (HAR) is essentially important to many applications, e.g., child/senior care, intelligent information push and exercise promotion. Although it is always better to build HAR model for each smart home to resolve the practical problem that homes have different floorplans or adopted sensors, it is intractable to acquire labeled data for each home due to cost and privacy. We thus propose a method to transfer the HAR model from multiple labeled source homes to the unlabeled target home. Specifically, we first generate transferable representations for the sensors of these homes, based on which we build the HAR model using the data of labeled source homes. Then, we employ the built HAR model into the unlabeled target home. Experiment results on CASAS dataset illustrate that our proposed method outperforms baseline methods in general and also avoids potential negative transfer caused by using only one source home.

Nowadays, advertising (ad) deliveries are conducted in a targeted manner to improve their effectiveness and efficiency. However, human behavior data in ad-platforms such as Web browsing history is complex and contains a lot of “noise”. On the other hand, information in the advertiser's domain (e.g. e-commerce sites) seems to contain less noise (e.g. product browsing history) with respect to ad-targeting. We introduce a new denoising method for behavioral ad-targeting based on the idea of feature distribution alignment induced by the advertiser's domain. This denoised distribution induction can be achieved by employing domain adversarial training with stabilization techniques. We evaluate our model on real world data originating from an e-commerce site and an ad-platform. The results of an ablation study have demonstrated the advantage of utilizing an advertiser's domain for denoising human behavior data of an ad-platform domain.

Behavioral targeting (BT) is an important function for a company to reach a wide range of potential users. Trans-domain BT which targets potential users of one (source) domain (e.g. E-Commerce) who lie in another (target) domain (e.g. Ad Network) is a promising method to expand the range. However, it is difficult for trans-domain BT to keep its targeting quality high in case when ID linkage across domains is limited. To realize high quality trans-domain BT with limited ID linkage, we propose a method to cross-connect private touchpoints to users in each domain, which we call Virtual Touch-Point (VTP). Here, we utilize transfer learning to acquire knowledge to tie two domains. We made a VTP prototype by implementing typical transfer learning algorithms and evaluated its effectiveness using real-world data of two domains: (source) E-Commerce → (target) Ad Network.