arXiv subjects

Ruijie Ma

Publications and source records attributed to Ruijie Ma.

5 recordsLinked to original sources

arXiv2025

Delving into Cryptanalytic Extraction of PReLU Neural Networks

The machine learning problem of model extraction was first introduced in 1991 and gained prominence as a cryptanalytic challenge starting with Crypto 2020. For over three decades, research in this field has primarily focused on ReLU-based neural networks. In this work, we take the first step towards the cryptanalytic extraction of PReLU neural networks, which employ more complex nonlinear activation functions than their ReLU counterparts. We propose a raw output-based parameter recovery attack for PReLU networks and extend it to more restrictive scenarios where only the top-m probability scores are accessible. Our attacks are rigorously evaluated through end-to-end experiments on diverse PReLU neural networks, including models trained on the MNIST dataset. To the best of our knowledge, this is the first practical demonstration of PReLU neural network extraction across three distinct attack scenarios.

cs.CR↗

arXiv2024

Enhancing Industrial Transfer Learning with Style Filter: Cost Reduction and Defect-Focus

Addressing the challenge of data scarcity in industrial domains, transfer learning emerges as a pivotal paradigm. This work introduces Style Filter, a tailored methodology for industrial contexts. By selectively filtering source domain data before knowledge transfer, Style Filter reduces the quantity of data while maintaining or even enhancing the performance of transfer learning strategy. Offering label-free operation, minimal reliance on prior knowledge, independence from specific models, and re-utilization, Style Filter is evaluated on authentic industrial datasets, highlighting its effectiveness when employed before conventional transfer strategies in the deep learning domain. The results underscore the effectiveness of Style Filter in real-world industrial applications.

cs.LG↗

arXiv2023

A method of weak lensing reconstruction through cosmic magnification with multi-band photometry information

Weak gravitational lensing induces flux dependent fluctuations in the observed galaxy number density distribution. This cosmic magnification (magnification bias) effect in principle enables lensing reconstruction alternative to cosmic shear and CMB lensing. However, the intrinsic galaxy clustering, which otherwise overwhelms the signal, has hindered its application. Through a scaling relation found by principal component analysis of the galaxy clustering in multi-band photometry space, we design a minimum variance linear estimator to suppress the intrinsic galaxy clustering and to reconstruct the lensing convergence map. In combination of the CosmoDC2 galaxy mock and the CosmicGrowth simulation, we test this proposal for a LSST-like galaxy survey with $ugrizY$ photometry bands. The scaling relation holds excellently at multipole $\ell<10^3$, and remains reasonably well to $\ell\sim 3000$. The linear estimator efficiently suppresses the galaxy intrinsic clustering, by a factor of $\sim 10^2$. For galaxies in the photo-z range $0.8 = 200$. Its cross-correlation with cosmic shear of galaxies can achieve $S/N >= 200$. When the source redshift of cosmic shear galaxies $z_γ 10^3$. This work further demonstrates the potential of lensing measurement through cosmic magnification to enhance the weak lensing cosmology.

astro-ph.CO↗

arXiv2023

Weak Lensing Reconstruction by Counting DECaLS Galaxies

Alternative to weak lensing measurements through cosmic shear, we present a weak lensing convergence $\hat{\kappa}$ map reconstructed through cosmic magnification effect in DECaLS galaxies of the DESI imaging surveys DR9. This is achieved by linearly weighing $12$ maps of galaxy number overdensity in different magnitude bins of $grz$ photometry bands. The weight is designed to eliminate the mean galaxy deterministic bias, minimize galaxy shot noise while maintaining the lensing convergence signal. We also perform corrections of imaging systematics in the galaxy number overdensity. The $\hat{\kappa}$ map has $8365$ deg$^2$ sky coverage. Given the low number density of DECaLS galaxies, the $\hat{\kappa}$ map is overwhelmed by shot noise and the map quality is difficult to evaluate using the lensing auto-correlation. Alternatively, we measure its cross-correlation with the cosmic shear catalogs of DECaLS galaxies of DESI imaging surveys DR8, which has $8365$ deg$^2$ overlap in sky coverage with the $\hat{\kappa}$ map. We detect a convergence-shear cross-correlation signal with $S/N\simeq 10$. The analysis also shows that the galaxy intrinsic clustering is suppressed by a factor $\mathcal{O}(10^2)$ and the residual galaxy clustering contamination in the $\hat{\kappa}$ map is consistent with zero. Various tests with different galaxy and shear samples, and the Akaike information criterion analysis all support the lensing detection. So is the imaging systematics corrections, which enhance the lensing signal detection by $\sim 30\%$. We discuss various issues for further improvement of the measurements.

astro-ph.CO↗

arXiv2023

BAD-NeRF: Bundle Adjusted Deblur Neural Radiance Fields

Neural Radiance Fields (NeRF) have received considerable attention recently, due to its impressive capability in photo-realistic 3D reconstruction and novel view synthesis, given a set of posed camera images. Earlier work usually assumes the input images are of good quality. However, image degradation (e.g. image motion blur in low-light conditions) can easily happen in real-world scenarios, which would further affect the rendering quality of NeRF. In this paper, we present a novel bundle adjusted deblur Neural Radiance Fields (BAD-NeRF), which can be robust to severe motion blurred images and inaccurate camera poses. Our approach models the physical image formation process of a motion blurred image, and jointly learns the parameters of NeRF and recovers the camera motion trajectories during exposure time. In experiments, we show that by directly modeling the real physical image formation process, BAD-NeRF achieves superior performance over prior works on both synthetic and real datasets. Code and data are available at https://github.com/WU-CVGL/BAD-NeRF.

cs.CV↗