Graph-based methods coupled with specific distributional distances for adversarial attack detection.

Artificial neural networks are prone to being fooled by carefully perturbed inputs which cause an egregious misclassification. These adversarial attacks have been the focus of extensive research. Likewise, there has been an abundance of research in ways to detect and defend against them. We introduce a novel approach of detection and interpretation of adversarial attacks from a graph perspective. For an input image, we compute an associated sparse graph using the layer-wise relevance propagation algorithm (Bach et al., 2015). Specifically, we only keep edges of the neural network with the highest relevance values. Three quantities are then computed from the graph which are then compared against those computed from the training set. The result of the comparison is a classification of the image as benign or adversarial. To make the comparison, two classification methods are introduced: (1) an explicit formula based on Wasserstein distance applied to the degree of node and (2) a logistic regression. Both classification methods produce strong results which lead us to believe that a graph-based interpretation of adversarial attacks is valuable.

Derivation and study of human epithelial cell lines resistant to killing by chromium trioxide.

CrO3 is cytotoxic for human epithelial 293 kidney cells over a narrow concentration range of approximately 2-8 microM (D50 approximately 3.0 microM); significantly greater toxicity is observed in clonogenic assays (D50 approximately 0.1-1.0 microM). Survival of a small fraction of cells (< or = 0.1%) at CrO3 concentrations between 10(-5) to 10(-3) M, and first-order kinetics of cytotoxicity, rationalized the derivation of a new panel of transformed human epithelial cell lines resistant to cytotoxic concentrations of CrO3 over the range of 5-100 microM. Wild-type and Cr-resistant 293 cell lines display similar morphology under phase microscopy, but wild-type cells grow faster and reach stationary phase sooner than Cr-resistant cells. The Cr-resistant phenotype is stable, and it is specific, since Cr-resistant cells are killed by NiSO4 or by CdCl2 at concentrations equivalent to those that kill wild-type cells. Toxicity analysis curves subjected to target theory suggest that the Cr-resistant cell lines have fewer Cr-sensitive "targets" and have undergone a "loss of function" compared to wild-type cells. This loss of function may be related to significantly lower rates of uptake of Na2(51)CrO4,which correlate inversely with CrO3 concentrations used for the selection and maintenance of the Cr-resistant lines, and to reduced levels of an approximately 96-kDa protein in comparison to wild-type cells. This new panel of Cr-resistant transformed human epithelial kidney cell lines will be useful in comparative studies of genetic resistance and sensitivity to human Cr(VI) toxicity, sulfate transport, and growth control differences between wild-type and Cr-resistant cells.