Malware is constantly adapting in order to avoid detection. Model based malware detectors, such as SVM and neural networks, are vulnerable to so-called adversarial examples which are modest changes to detectable malware that allows the resulting malware to evade detection. Continuous-valued methods that are robust to adversarial examples of images have been developed using saddle-point optimization formulations. We are inspired by them to develop similar methods for the discrete, e.g. binary, domain which characterizes the features of malware. A specific extra challenge of malware is that the adversarial examples must be generated in a way that preserves their malicious functionality. We introduce methods capable of generating functionally preserved adversarial malware examples in the binary domain. Using the saddle-point formulation, we incorporate the adversarial examples into the training of models that are robust to them. We evaluate the effectiveness of the methods and others in the literature on a set of Portable Execution (PE) files. Comparison prompts our introduction of an online measure computed during training to assess general expectation of robustness.

Credit cards have often been blamed for consumer overspending and for the growth in household debt. Indeed, laboratory studies of purchase behavior have shown that credit cards can facilitate spending in ways that are difficult to justify on purely financial grounds. However, the psychological mechanisms behind this spending facilitation effect remain conjectural. A leading hypothesis is that credit cards reduce the pain of payment and so ‘release the brakes’ that hold expenditures in check. Alternatively, credit cards could provide a ‘step on the gas,’ increasing motivation to spend. Here we present the first evidence of differences in brain activation in the presence of real credit and cash purchase opportunities. In an fMRI shopping task, participants purchased items tailored to their interests, either by using a personal credit card or their own cash. Credit card purchases were associated with strong activation in the striatum, which coincided with onset of the credit card cue and was not related to product price. In contrast, reward network activation weakly predicted cash purchases, and only among relatively cheaper items. The presence of reward network activation differences highlights the potential neural impact of novel payment instruments in stimulating spending—these fundamental reward mechanisms could be exploited by new payment methods as we transition to a purely cashless society.

No abstract

Numerical simulations have been carried out to define the loss generation mechanisms associated with tip leakage in unshrouded axial turbines. Tip clearance vortex dynamics are a dominant feature of two mechanisms important in determining this loss: (i) decreased swirl velocity due to vortex line contraction in regions of decreasing axial velocity, i.e., adverse pressure gradient, and (ii) vortex breakdown and reverse flow in the vortex core. The mixing losses behave differently from the conventional view of flow exiting a turbine tip clearance. More specifically, it is shown, through control volume arguments and computations, that as a swirling leakage flow passes through a pressure rise, such as in the aft portion of the suction side of a turbine blade, the mixed-out loss can either decrease or increase. For turbines the latter typically occurs if the deceleration is large enough to initiate vortex breakdown, and it is demonstrated that this can occur case in modern turbines. The effect of blade pressure distribution on clearance losses is illustrated through computational examination of turbine blades with forward loading at the tip and with aft loading. A 15% difference in leakage loss is found between the two, due to lower clearance vortex deceleration (lower core static pressure rise) with forward loading and hence lower vortex breakdown loss. Additional computational experiments, carried out to define the effects of blade loading, incidence, and solidity, are found to be consistent with the proposed ideas linking blade pressure distribution, vortex breakdown and turbine tip leakage loss. NOMENCLATURE AR Duct exit-to-inlet area ratio INTRODUCTIONThe leakage flow through the radial clearance between the tip of the rotating blades and the stationary casing is an important turbomachinery loss source. For an axial turbine, the decrease in efficiency from this loss is roughly proportional to clearance/span for a given design, [1], with proportionality levels between 1 to 3. Typical clearances for axial turbines lie in the range of 1-2% clearance/span [2] so the clearance loss for an unshrouded turbine is about one-third of the total loss [1,3].The characteristics of clearance flow over flat-tipped blades have been well-described [1,4]. Flow enters on the pressure side. With a sharp corner a separation bubble forms on the tip at the pressure side corner. If the blade thickness to clearance ratio is greater than approximately 2, the flow will reattach [5] before exiting the clearance gap. At the gap exit the clearance flow and free stream velocities are not aligned and a shear layer (vortex sheet) exists between clearance and mainstream flows, which rolls up into a vortex.A standard approach in describing the leakage flow is to divide the process into two fluid dynamic modules [6][7][8]. The first considers the flow passing through the gap. The second considers the interaction between the leakage flow and the mainstream.For the first module the stagnation-to-static pressure ratio, including any losses, d...

A central challenge of adversarial learning is to interpret the resulting hardened model. In this contribution, we ask how robust generalization can be visually discerned and whether a concise view of the interactions between a hardened decision map and input samples is possible. We first provide a means of visually comparing a hardened model's loss behavior with respect to the adversarial variants generated during training versus loss behavior with respect to adversarial variants generated from other sources. This allows us to confirm that the association of observed flatness of a loss landscape with generalization that is seen with naturally trained models extends to adversarially hardened models and robust generalization. To complement these means of interpreting model parameter robustness we also use self-organizing maps to provide a visual means of superimposing adversarial and natural variants on a model's decision space, thus allowing the model's global robustness to be comprehensively examined.