Search for: All records

We propose AccHashtag, the first framework for high-accuracy detection of fault-injection attacks on Deep Neural Networks (DNNs) with provable bounds on detection performance. Recent literature in fault-injection attacks shows the severe DNN accuracy degradation caused by bit flips. In this scenario, the attacker changes a few DNN weight bits during execution by injecting faults to the dynamic random-access memory (DRAM). To detect bit flips, AccHashtag extracts a unique signature from the benign DNN prior to deployment. The signature is used to validate the model’s integrity and verify the inference output on the fly. We propose a novel sensitivity analysis that identifies the most vulnerable DNN layers to the fault-injection attack. The DNN signature is constructed by encoding the weights in vulnerable layers using a low-collision hash function. During DNN inference, new hashes are extracted from the target layers and compared against the ground-truth signatures. AccHashtag incorporates a lightweight methodology that allows for real-time fault detection on embedded platforms. We devise a specialized compute core for AccHashtag on field-programmable gate arrays (FPGAs) to facilitate online hash generation in parallel to DNN execution. Extensive evaluations with the state-of-the-art bit-flip attack on various DNNs demonstrate the competitive advantage of AccHashtag in terms of both attack detection and execution overhead. 

The effect of catalyst addition on H2 evolution from composite electrodes of La0.7Sr0.3TiO3 (LST) and BaZr0.1Ce0.7Y0.1Yb0.1O3−δ (BZCYYb) was studied. Starting with symmetric cells (LST∣∣BZCYYb∣∣LST), Pt was added to one or both electrodes, after which i–V polarization measurements were performed in humidified H2 at 723 and 773 K. The base cells showed very high impedances but these decreased dramatically upon addition of Pt to both electrodes. When Pt was added to only one electrode, the cells performed as diodes, showing that Pt was necessary for H2 dissociation but not for H recombination. The effects of adding Ru, W, Re and Fe were also studied. DFT calculations helped confirm that H recombination on BaZrO3 is expected to be barrierless. The implications of these results for potential application to electrochemical synthesis of ammonia are discussed. 

ABSTRACT We perform analysis of the 3D kinematics of Milky Way disc stars in mono-age populations. We focus on stars between Galactocentric distances of R = 6 and 14  kpc, selected from the combined LAMOST Data Release 4 (DR4) red clump giant stars and Gaia DR2 proper motion catalogue. We confirm the 3D asymmetrical motions of recent works and provide time tagging of the Galactic outer disc asymmetrical motions near the anticentre direction out to Galactocentric distances of 14 kpc. Radial Galactocentric motions reach values up to 10 km s−1, depending on the age of the population, and present a north–south asymmetry in the region corresponding to density and velocity substructures that were sensitive to the perturbations in the early 6  Gyr. After that time, the disc stars in this asymmetrical structure have become kinematically hotter, and are thus not sensitive to perturbations, and we find the structure is a relatively younger population. With quantitative analysis, we find stars both above and below the plane at R ≳ 9 kpc that exhibit bending mode motions of which the sensitive duration is around 8  Gyr. We speculate that the in-plane asymmetries might not be mainly caused by a fast rotating bar, intrinsically elliptical outer disc, secular expansion of the disc, or streams. Spiral arm dynamics, out-of-equilibrium models, minor mergers or others are important contributors. Vertical motions might be dominated by bending and breathing modes induced by complicated inner or external perturbers. It is likely that many of these mechanisms are coupled together.