Boosting Factorization Machines via Saliency-Guided Mixup.

Factorization machines (FMs) are widely used in recommender systems due to their adaptability and ability to learn from sparse data. However, for the ubiquitous non-interactive features in sparse data, existing FMs can only estimate the parameters corresponding to these features via the inner product of their embeddings. Undeniably, they cannot learn the direct interactions of these features, which limits the model's expressive power. To this end, we first present MixFM, inspired by Mixup, to generate auxiliary training data to boost FMs. Unlike existing augmentation strategies that require labor costs and expertise to collect additional information such as position and fields, these augmented data are only by the convex combination of the raw ones without any professional knowledge support. More importantly, if non-interactive features exist in parent samples to be mixed respectively, MixFM will establish their direct interactions. Second, considering that MixFM may generate redundant or even detrimental instances, we further put forward a novel Factorization Machine powered by Saliency-guided Mixup (denoted as SMFM). Guided by the customized saliency, SMFM can generate more informative neighbor data. Through theoretical analysis, we prove that the proposed methods minimize the upper bound of the generalization error, which positively enhances FMs. Finally, extensive experiments on seven datasets confirm that our approaches are superior to baselines. Notably, the results also show that "poisoning" mixed data benefits the FM variants.

Influence-Driven Data Poisoning for Robust Recommender Systems.

Recent studies have shown that recommender systems are vulnerable, and it is easy for attackers to inject well-designed malicious profiles into the system, resulting in biased recommendations. We cannot deprive these data's injection right and deny their existence's rationality, making it imperative to study recommendation robustness. Despite impressive emerging work, threat assessment of the bi-level poisoning problem and the imperceptibility of poisoning users remain key challenges to be solved. To this end, we propose Infmix, an efficient poisoning attack strategy. Specifically, Infmix consists of an influence-based threat estimator and a user generator, Usermix. First, the influence-based estimator can efficiently evaluate the user's harm to the recommender system without retraining, which is challenging for existing attacks. Second, Usermix, a distribution-agnostic generator, can generate unnoticeable fake data even with a few known users. Under the guidance of the threat estimator, Infmix can select the users with large attacking impacts from the quasi-real candidates generated by Usermix. Extensive experiments demonstrate Infmix's superiority by attacking six recommendation systems with four real datasets. Additionally, we propose a novel defense strategy, adversarial poisoning training (APT). It mimics the poisoning process by injecting fake users (ERM users) committed to minimizing empirical risk to build a robust system. Similar to Infmix, we also utilize the influence function to solve the bi-level optimization challenge of generating ERM users. Although the idea of "fighting fire with fire" in APT seems counterintuitive, we prove its effectiveness in improving recommendation robustness through theoretical analysis and empirical experiments.

MAD-Net: Multi-attention dense network for functional bone marrow segmentation.

Radiotherapy is the main treatment modality for various pelvic malignancies. However, high intensity radiation can damage the functional bone marrow (FBM), resulting in hematological toxicity (HT). Accurate identification and protection of the FBM during radiotherapy planning can reduce pelvic HT. The traditional manual method for contouring the FBM is time-consuming and laborious. Therefore, development of an efficient and accurate automatic segmentation mode can provide a distinct leverage in clinical settings. In this paper, we propose the first network for performing the FBM segmentation task, which is referred to as the multi-attention dense network (named MAD-Net). Primarily, we introduce the dense convolution block to promote the gradient flow in the network as well as incite feature reuse. Next, a novel slide-window attention module is proposed to emphasize long-range dependencies and exploit interdependencies between features. Finally, we design a residual-dual attention module as the bottleneck layer, which further aggregates useful spatial details and explores intra-class responsiveness of high-level features. In this work, we conduct extensive experiments on our dataset of 3838 two-dimensional pelvic slices. Experimental results demonstrate that the proposed MAD-Net transcends previous state-of-the-art models in various metrics. In addition, the contributions of the proposed components are verified by ablation analysis, and we conduct experiments on three other datasets to manifest the generalizability of MAD-Net.

Multiple classes of chemical contaminants in soil from an e-waste disposal site in China: Occurrence and spatial distribution.

E-waste recycling is well known for releasing halogenated organic compounds (HOCs) and heavy metals. This study investigated the occurrence and distribution of traditional and novel classes of contaminants, including chlorinated, brominated, and mixed halogenated dibenzo-p-dioxins/dibenzofurans (PCDD/Fs, PBDD/Fs, PXDD/Fs), polybrominated diphenyl ethers (PBDEs), polychlorinated biphenyls (PCBs) and polyhalogenated carbazoles (PHCZs), in soil from an e-waste disposal site in Hangzhou. PBDEs were the most abundant, at 343-69306 ng kg-1, followed by PHCZs (896-41,362 ng kg-1), PCDD/Fs (349-19,396 ng kg-1), PCBs (51.3-1834 ng kg-1), PBDD/Fs (2.99-524 ng kg-1) and PXDD/Fs (0.104-21.2 ng kg-1). The detected target compound concentrations were generally lower than those reported in the literature for informal e-waste sites. Nevertheless, they can serve as a basis of information for evaluation and subsequent control. The toxic equivalent (TEQ) contributions from these contaminants (except PBDEs) decreased as follows: PCDD/Fs > PXDD/Fs > PHCZs > PCBs > PBDD/Fs. ΣDioxins (PCDD/Fs + PBDD/Fs + PXDD/Fs) accounted for 47.7%-97.2% of the total TEQs in the soil. OCDD, 1,2,3,4,6,7,8-HpBDF and OBDF were the dominant congeners, mainly derived from combustion and transport because of their low saturated vapor pressure. PXDFs were more abundant than PXDDs, and homologue profiles suggested a similar formation mechanism for PXDFs and PBDFs involving successive Br-to-Cl exchange. PHCZs were reported in soil from an e-waste disposal area for the first time, and their concentrations were several orders of magnitude higher than those of the other contaminants. Although the risk of human exposure in this study was estimated to be lower than the values recommended by the WHO (1-4 pg TEQ/kg bw/day), health implications still exist, and further investigations are necessary.

Method development for analyzing ultratrace polyhalogenated carbazoles in soil and sediment.

In the past few years, polyhalogenated carbazoles (PHCZs) have been of increasing concern because their structure is similar to that of legacy POPs. In the present study, an analytical method, including intensive cleanup and fractionation procedures in combination with instrumental parameters, was developed to determine ultratrace polyhalogenated carbazoles (PHCZs) in soil and sediment. The eluting sorbents, volume and packing of the column were optimized. Our results showed that 5 g of florisil and 4 g of silica gel under 150 mL of hexane/DCM = 3:1 presented good performance in terms of recovery and repeatability. GC-HRMS, GC-MS/MS (EI-MRM) and GC-MS (EI-SIM) were applied to compare the performance of PHCZ analysis. For sensitivity, EI-MRM presents method detection limits comparable to those of GC-HRMS and much lower than those of EI-SIM. Regarding selectivity, GC-HRMS performed better than the other two techniques since GC-HRMS can reduce interference from perfluorokerosene (PFK) and DDX (DDT, DDE, and DDD) due to its high resolution. GC-HRMS was then further optimized by shortening the run time and modifying the SIM ion. The final method was successfully applied to determine PHCZs in soil and sediment, and the target compounds had almost 100% detection frequency in the samples. The ubiquitous presence of PHCZ in soil and sediment calls for a further investigation of its source, distribution and degradation in the environment.

Study on the pollution characteristics and emission factors of PCDD/Fs from disperse dye production in China.

According to an analysis of the input and output of polychlorinated dibenzo-p-dioxin and dibenzofurans (PCDD/Fs) of two disperse dye manufacturers, the average PCDD/F emission factor was 56.3 µg I-TEQ/T product, and dioxin was mainly emitted with products and solid waste. The PCDD/F concentrations of different series of disperse dyes varied from 2.87 to 323 pg I-TEQ/g, and the dominant congener was OCDD or 2,3,7,8-TCDF with the highest ratio of 83.4% or 79.3%, respectively. The distributions of PCDD/F congeners in raw materials and the products were different, indicating that the structure of PCDD/Fs greatly changed in the synthesis process. PCDD/Fs in the wastewater and sludge of the companies are mainly from the production process of a product with dominant emission factors. Our results confirm that disperse dyes may be a source of PCDD/Fs, resulting in human exposure and environmental contamination.