arXiv subjects

Ziang Yan

Publications and source records attributed to Ziang Yan.

64 records · Page 4Linked to original sources

arXiv2019

An Assessment of Contamination in the thermal-SZ Map Using Cross Correlations

We search for the potential contamination in the Planck thermal Sunyaev-Zeldovich (tSZ) map by calculating the cross-correlation between the tSZ signal and weak lensing by large scale structure and the Cosmic Microwave Background (CMB). The lensing data we use is the convergence map from the Red Sequence Cluster Lensing Survey (RCSLens) and the Planck CMB lensing map. We reconstruct the tSZ $y$ map with a Needlet Internal Linear Combination method using the HFI sky maps from the Planck satellite. We remove the CMB signal while minimizing the residual noise. The cross correlation signal from our reconstructed $y$ map is consistent with that from the Planck team's NILC $y$ map. The CIB and galactic dust emission are two potential sources of contamination in the reconstructed $y$ map. We remove the CIB signal by subtracting the CIB maps reconstructed by Planck collaboration from the raw temperature maps. We find that cross-correlation between the CIB and galactic lensing contributes to $(7.5\pm6.0)\%$ in the Planck NILC tSZ cross galactic lensing signal within $100<\ell<2000$, which implies that previous detections of the tSZ cross galactic lensing is robust to CIB contaminations. In contrast, the Planck NILC tSZ cross CMB lensing is biased by $(18.4\pm2.8)\%$ in the same $\ell$ range. Galactic dust contamination is tested by projecting out a grey-body dust models with different dust spectral indices. Galactic dust does not affect galactic lensing cross tSZ signal significantly.

astro-ph.CO↗

arXiv2019

Subspace Attack: Exploiting Promising Subspaces for Query-Efficient Black-box Attacks

Unlike the white-box counterparts that are widely studied and readily accessible, adversarial examples in black-box settings are generally more Herculean on account of the difficulty of estimating gradients. Many methods achieve the task by issuing numerous queries to target classification systems, which makes the whole procedure costly and suspicious to the systems. In this paper, we aim at reducing the query complexity of black-box attacks in this category. We propose to exploit gradients of a few reference models which arguably span some promising search subspaces. Experimental results show that, in comparison with the state-of-the-arts, our method can gain up to 2x and 4x reductions in the requisite mean and medium numbers of queries with much lower failure rates even if the reference models are trained on a small and inadequate dataset disjoint to the one for training the victim model. Code and models for reproducing our results will be made publicly available.

cs.CV↗

arXiv2018

Deep Defense: Training DNNs with Improved Adversarial Robustness

Despite the efficacy on a variety of computer vision tasks, deep neural networks (DNNs) are vulnerable to adversarial attacks, limiting their applications in security-critical systems. Recent works have shown the possibility of generating imperceptibly perturbed image inputs (a.k.a., adversarial examples) to fool well-trained DNN classifiers into making arbitrary predictions. To address this problem, we propose a training recipe named "deep defense". Our core idea is to integrate an adversarial perturbation-based regularizer into the classification objective, such that the obtained models learn to resist potential attacks, directly and precisely. The whole optimization problem is solved just like training a recursive network. Experimental results demonstrate that our method outperforms training with adversarial/Parseval regularizations by large margins on various datasets (including MNIST, CIFAR-10 and ImageNet) and different DNN architectures. Code and models for reproducing our results are available at https://github.com/ZiangYan/deepdefense.pytorch

cs.CV↗

arXiv2018

The Cosmology Large Angular Scale Surveyor Receiver Design

The Cosmology Large Angular Scale Surveyor consists of four instruments performing a CMB polarization survey. Currently, the 40 GHz and first 90 GHz instruments are deployed and observing, with the second 90 GHz and a multichroic 150/220 GHz instrument to follow. The receiver is a central component of each instrument's design and functionality. This paper describes the CLASS receiver design, using the first 90 GHz receiver as a primary reference. Cryogenic cooling and filters maintain a cold, low-noise environment for the detectors. We have achieved receiver detector temperatures below 50 mK in the 40 GHz instrument for 85% of the initial 1.5 years of operation, and observed in-band efficiency that is consistent with pre-deployment estimates. At 90 GHz, less than 26% of in-band power is lost to the filters and lenses in the receiver, allowing for high optical efficiency. We discuss the mounting scheme for the filters and lenses, the alignment of the cold optics and detectors, stray light control, and magnetic shielding.

astro-ph.IM↗

arXiv2018

Design and characterization of the Cosmology Large Angular Scale Surveyor (CLASS) 93 GHz focal plane

The Cosmology Large Angular Scale Surveyor (CLASS) aims to detect and characterize the primordial B-mode signal and make a sample-variance-limited measurement of the optical depth to reionization. CLASS is a ground-based, multi-frequency microwave polarimeter that surveys 70% of the microwave sky every day from the Atacama Desert. The focal plane detector arrays of all CLASS telescopes contain smooth-walled feedhorns that couple to transition-edge sensor (TES) bolometers through symmetric planar orthomode transducer (OMT) antennas. These low noise polarization-sensitive detector arrays are fabricated on mono-crystalline silicon wafers to maintain TES uniformity and optimize optical efficiency throughout the wafer. In this paper, we discuss the design and characterization of the first CLASS 93 GHz detector array. We measure the dark parameters, bandpass, and noise spectra of the detectors and report that the detectors are photon-noise limited. With current array yield of 82%, we estimate the total array noise-equivalent power (NEP) to be 2.1 aW$\sqrt[]{\mathrm{s}}$.

astro-ph.IM↗

arXiv2018

Variable-delay Polarization Modulators for the CLASS Telescopes

The search for inflationary primordial gravitational waves and the measurement of the optical depth to reionization, both through their imprint on the large angular scale correlations in the polarization of the cosmic microwave background (CMB), has created the need for high sensitivity measurements of polarization across large fractions of the sky at millimeter wavelengths. These measurements are subject to instrumental and atmospheric $1/f$ noise, which has motivated the development of polarization modulators to facilitate the rejection of these large systematic effects. Variable-delay polarization modulators (VPMs) are used in the Cosmology Large Angular Scale Surveyor (CLASS) telescopes as the first element in the optical chain to rapidly modulate the incoming polarization. VPMs consist of a linearly polarizing wire grid in front of a movable flat mirror. Varying the distance between the grid and the mirror produces a changing phase shift between polarization states parallel and perpendicular to the grid which modulates Stokes U (linear polarization at $45^\circ$) and Stokes V (circular polarization). The CLASS telescopes have VPMs as the first optical element from the sky; this simultaneously allows a lock-in style polarization measurement and the separation of sky polarization from any instrumental polarization further along in the optical path. The Q-band CLASS VPM was the first VPM to begin observing the CMB full time, starting in the Spring of 2016. The first W-band CLASS VPM was installed in the Spring of 2018.

astro-ph.IM↗

arXiv2018

Optimizing Recurrent Neural Networks Architectures under Time Constraints

Recurrent neural network (RNN)'s architecture is a key factor influencing its performance. We propose algorithms to optimize hidden sizes under running time constraint. We convert the discrete optimization into a subset selection problem. By novel transformations, the objective function becomes submodular and constraint becomes supermodular. A greedy algorithm with bounds is suggested to solve the transformed problem. And we show how transformations influence the bounds. To speed up optimization, surrogate functions are proposed which balance exploration and exploitation. Experiments show that our algorithms can find more accurate models or faster models than manually tuned state-of-the-art and random search. We also compare popular RNN architectures using our algorithms.

stat.ML↗

arXiv2018

Neural Network Architecture Optimization through Submodularity and Supermodularity

Deep learning models' architectures, including depth and width, are key factors influencing models' performance, such as test accuracy and computation time. This paper solves two problems: given computation time budget, choose an architecture to maximize accuracy, and given accuracy requirement, choose an architecture to minimize computation time. We convert this architecture optimization into a subset selection problem. With accuracy's submodularity and computation time's supermodularity, we propose efficient greedy optimization algorithms. The experiments demonstrate our algorithm's ability to find more accurate models or faster models. By analyzing architecture evolution with growing time budget, we discuss relationships among accuracy, time and architecture, and give suggestions on neural network architecture design.

stat.ML↗

arXiv2017

Zero-Shot Learning by Generating Pseudo Feature Representations

Zero-shot learning (ZSL) is a challenging task aiming at recognizing novel classes without any training instances. In this paper we present a simple but high-performance ZSL approach by generating pseudo feature representations (GPFR). Given the dataset of seen classes and side information of unseen classes (e.g. attributes), we synthesize feature-level pseudo representations for novel concepts, which allows us access to the formulation of unseen class predictor. Firstly we design a Joint Attribute Feature Extractor (JAFE) to acquire understandings about attributes, then construct a cognitive repository of attributes filtered by confidence margins, and finally generate pseudo feature representations using a probability based sampling strategy to facilitate subsequent training process of class predictor. We demonstrate the effectiveness in ZSL settings and the extensibility in supervised recognition scenario of our method on a synthetic colored MNIST dataset (C-MNIST). For several popular ZSL benchmark datasets, our approach also shows compelling results on zero-shot recognition task, especially leading to tremendous improvement to state-of-the-art mAP on zero-shot retrieval task.

cs.CV↗

arXiv2017

Weakly- and Semi-Supervised Object Detection with Expectation-Maximization Algorithm

Object detection when provided image-level labels instead of instance-level labels (i.e., bounding boxes) during training is an important problem in computer vision, since large scale image datasets with instance-level labels are extremely costly to obtain. In this paper, we address this challenging problem by developing an Expectation-Maximization (EM) based object detection method using deep convolutional neural networks (CNNs). Our method is applicable to both the weakly-supervised and semi-supervised settings. Extensive experiments on PASCAL VOC 2007 benchmark show that (1) in the weakly supervised setting, our method provides significant detection performance improvement over current state-of-the-art methods, (2) having access to a small number of strongly (instance-level) annotated images, our method can almost match the performace of the fully supervised Fast RCNN. We share our source code at https://github.com/ZiangYan/EM-WSD.

cs.CV↗