Source author record

Bryan Kim

Explore this author’s publications, collaborators and research topics. This catalog record has not been claimed by its author.

ResearcherUnclaimed source record

astro-ph astro-ph.IM Cryptography and Security Machine Learning

Catalog footprint

Research overview

3works

4topics

4close collaborators

Actions

Is this your work?

Open graph Browse papers

preprint2020arXiv

Investigating a Deep Learning Method to Analyze Images from Multiple Gamma-ray Telescopes

Imaging atmospheric Cherenkov telescope (IACT) arrays record images from air showers initiated by gamma rays entering the atmosphere, allowing astrophysical sources to be observed at very high energies. To maximize IACT sensitivity, gamma-ray showers must be efficiently distinguished from the dominant background of cosmic-ray showers using images from multiple telescopes. A combination of convolutional neural networks (CNNs) with a recurrent neural network (RNN) has been proposed to perform this task. Using CTLearn, an open source Python package using deep learning to analyze data from IACTs, with simulated data from the upcoming Cherenkov Telescope Array (CTA), we implement a CNN-RNN network and find no evidence that sorting telescope images by total amplitude improves background rejection performance.

preprint2020arXiv

Live Trojan Attacks on Deep Neural Networks

Like all software systems, the execution of deep learning models is dictated in part by logic represented as data in memory. For decades, attackers have exploited traditional software programs by manipulating this data. We propose a live attack on deep learning systems that patches model parameters in memory to achieve predefined malicious behavior on a certain set of inputs. By minimizing the size and number of these patches, the attacker can reduce the amount of network communication and memory overwrites, with minimal risk of system malfunctions or other detectable side effects. We demonstrate the feasibility of this attack by computing efficient patches on multiple deep learning models. We show that the desired trojan behavior can be induced with a few small patches and with limited access to training data. We describe the details of how this attack is carried out on real systems and provide sample code for patching TensorFlow model parameters in Windows and in Linux. Lastly, we present a technique for effectively manipulating entropy on perturbed inputs to bypass STRIP, a state-of-the-art run-time trojan detection technique.

preprint2005arXiv

The velocity field of baryonic gas in the universe

The dynamic evolution of the baryonic intergalactic medium (IGM) caused by the underlying dark matter gravity is governed by the Navier-Stokes equations in which many cooling and heating processes are involved. However, it has long been recognized that the growth mode dynamics of cosmic matter clustering can be sketched by a random force driven Burgers' equation if cooling and heating are ignored. Just how well the dynamics of the IGM can be described as a Burgers fluid has not been fully investigated probably because cooling and heating are essential for a detailed understanding of the IGM. Using IGM samples produced by a cosmological hydrodynamic simulation in which heating and cooling processes are properly accounted for, we show that the IGM velocity field in the nonlinear regime shows the features of a Burgers fluid, that is, when the Reynolds number is high, the velocity field consists of an ensemble of shocks. Consequently, (1) the IGM velocity $v$ is generally smaller than that of dark matter; (2) for the smoothed field, the IGM velocity shows tight correlation with dark matter given by $v \simeq s v_{dm}$ , with $s<1$ , such that the lower the redshift, the smaller $s$ ; (3) the velocity PDFs are asymmetric between acceleration and deceleration events; (4) the PDF of velocity difference $Δv=v(x+r)-v(x)$ satisfies the scaling relation for a Burgers fluid, i.e., $P(Δv)=(1 r^y)F(Δv/r^y)$ . We find the scaling function and parameters for the IGM which are applicable to the entire scale range of the samples (0.26 - 8 h $^{-1}$ Mpc). These properties show that the similarity mapping between the IGM and dark matter is violated on scales much larger than the Jeans length of the IGM.

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.