Deep Neural Networks (DNNs) are vulnerable to Neural Trojan (NT) attacks where the adversary injects malicious behaviors during DNN training. This type of ‘backdoor’ attack is activated when the input is stamped with the trigger pattern specified by the attacker, resulting in an incorrect prediction of the model. Due to the wide application of DNNs in various critical fields, it is indispensable to inspect whether the pre-trained DNN has been trojaned before employing a model. Our goal in this paper is to address the security concern on unknown DNN to NT attacks and ensure safe model deployment. We propose DeepInspect, the first black-box Trojan detection solution with minimal prior knowledge of the model. DeepInspect learns the probability distribution of potential triggers from the queried model using a conditional generative model, thus retrieves the footprint of backdoor insertion. In addition to NT detection, we show that DeepInspect’s trigger generator enables effective Trojan mitigation by model patching. We corroborate the effectiveness, efficiency, and scalability of DeepInspect against the state-of-the-art NT attacks across various benchmarks. Extensive experiments show that DeepInspect offers superior detection performance and lower runtime overhead than the prior work.

Cell adhesion, mediated by specific receptor-ligand interactions, plays an important role in biological processes such as tumor metastasis and inflammatory cascade. For example, interactions between beta 2-integrin (lymphocyte function-associated antigen-1 and/or Mac-1) on polymorphonuclear neutrophils (PMNs) and ICAM-1 on melanoma cells initiate the bindings of melanoma cells to PMNs within the tumor microenvironment in blood flow, which in turn activate PMN-melanoma cell aggregation in a near-wall region of the vascular endothelium, therefore enhancing subsequent extravasation of melanoma cells in the microcirculations. Kinetics of integrin-ligand bindings in a shear flow is the determinant of such a process, which has not been well understood. In the present study, interactions of PMNs with WM9 melanoma cells were investigated to quantify the kinetics of beta 2-integrin and ICAM-1 bindings using a cone-plate viscometer that generates a linear shear flow combined with a two-color flow cytometry technique. Aggregation fractions exhibited a transition phase where it first increased before 60 s and then decreased with shear durations. Melanoma-PMN aggregation was also found to be inversely correlated with the shear rate. A previously developed probabilistic model was modified to predict the time dependence of aggregation fractions at different shear rates and medium viscosities. Kinetic parameters of beta 2-integrin and ICAM-1 bindings were obtained by individual or global fittings, which were comparable to respectively published values. These findings provide new quantitative understanding of the biophysical basis of leukocyte-tumor cell interactions mediated by specific receptor-ligand interactions under shear flow conditions.

Cancer metastasis involves multistep adhesive interactions between tumor cells (TCs) and endothelial cells (ECs), but the molecular mechanisms of intercellular communication in the tumor microenvironment remain elusive. Using static and flow coculture systems in conjunction with flow cytometry, we discovered that certain receptors on the ECs are upregulated on melanoma cell adhesion. Direct contact but not separate coculture between human umbilical endothelial cells (HUVECs) and a human melanoma cell line (Lu1205) increased intercellular adhesion molecule 1 (ICAM-1) and E-selectin expression on HUVECs by 3- and 1.5-fold, respectively, compared with HUVECs alone. The nonmetastatic cell line WM35 failed to promote ICAM-1 expression changes in HUVECs on contact. Enzyme-linked immunosorbent assay (ELISA) revealed that EC-TC contact has a synergistic effect on the expression of the cytokines interleukin (IL)-8, IL-6, and growth-related oncogene α (Gro-α). By using E-selectin cross-linking and beads coated with CD44 immunopurified from Lu1205 cells, we showed that CD44/selectin ligation was responsible for the ICAM-1 up-regulation on HUVECs. Protein kinase Cα (PKC-α) activation was found to be the downstream target of the CD44/selectin-initiated signaling, as ICAM-1 elevation was inhibited by siRNA targeting PKCα or a dominant negative form of PKCα (PKCα DN). Western blot analysis and electrophoretic mobility shift assays (EMSAs) showed that TC-EC contact mediated p38 phosphorylation and binding of the transcription factor SP-1 to its regulation site. In conclusion, CD44/selectin binding signals ICAM-1 up-regulation on the EC surface through a PKCα-p38-SP-1 pathway, which further enhances melanoma cell adhesion to ECs during metastasis.

Entity resolution (ER) aims to identify data records referring to the same real-world entity. Due to the heterogeneity of entity attributes and the diversity of similarity measures, one main challenge of ER is how to select appropriate similarity measures for different attributes. Previous ER methods usually employ heuristic similarity selection algorithms, which are highly specialized to specific ER problems and are hard to be generalized to other situations. Furthermore, previous studies usually perform similarity learning and similarity selection independently, which often result in error propagation and are hard to be optimized globally. To resolve the above problems, this paper proposes an end-to-end multi-perspective entity matching model, which can adaptively select optimal similarity measures for heterogenous attributes by jointly learning and selecting similarity measures in an end-to-end way. Experiments on two real-world datasets show that our method significantly outperforms previous ER methods.