arXiv subjects

Changxi Zheng

Publications and source records attributed to Changxi Zheng.

52 records · Page 3Linked to original sources

arXiv2019

One Man's Trash is Another Man's Treasure: Resisting Adversarial Examples by Adversarial Examples

Modern image classification systems are often built on deep neural networks, which suffer from adversarial examples--images with deliberately crafted, imperceptible noise to mislead the network's classification. To defend against adversarial examples, a plausible idea is to obfuscate the network's gradient with respect to the input image. This general idea has inspired a long line of defense methods. Yet, almost all of them have proven vulnerable. We revisit this seemingly flawed idea from a radically different perspective. We embrace the omnipresence of adversarial examples and the numerical procedure of crafting them, and turn this harmful attacking process into a useful defense mechanism. Our defense method is conceptually simple: before feeding an input image for classification, transform it by finding an adversarial example on a pre-trained external model. We evaluate our method against a wide range of possible attacks. On both CIFAR-10 and Tiny ImageNet datasets, our method is significantly more robust than state-of-the-art methods. Particularly, in comparison to adversarial training, our method offers lower training cost as well as stronger robustness.

cs.LG↗

arXiv2019

Enhancing Adversarial Defense by k-Winners-Take-All

We propose a simple change to existing neural network structures for better defending against gradient-based adversarial attacks. Instead of using popular activation functions (such as ReLU), we advocate the use of k-Winners-Take-All (k-WTA) activation, a C0 discontinuous function that purposely invalidates the neural network model's gradient at densely distributed input data points. The proposed k-WTA activation can be readily used in nearly all existing networks and training methods with no significant overhead. Our proposal is theoretically rationalized. We analyze why the discontinuities in k-WTA networks can largely prevent gradient-based search of adversarial examples and why they at the same time remain innocuous to the network training. This understanding is also empirically backed. We test k-WTA activation on various network structures optimized by a training method, be it adversarial training or not. In all cases, the robustness of k-WTA networks outperforms that of traditional networks under white-box attacks.

cs.LG↗

arXiv2019

Rethinking Generative Mode Coverage: A Pointwise Guaranteed Approach

Many generative models have to combat $\textit{missing modes}$. The conventional wisdom to this end is by reducing through training a statistical distance (such as $f$-divergence) between the generated distribution and provided data distribution. But this is more of a heuristic than a guarantee. The statistical distance measures a $\textit{global}$, but not $\textit{local}$, similarity between two distributions. Even if it is small, it does not imply a plausible mode coverage. Rethinking this problem from a game-theoretic perspective, we show that a complete mode coverage is firmly attainable. If a generative model can approximate a data distribution moderately well under a global statistical distance measure, then we will be able to find a mixture of generators that collectively covers $\textit{every}$ data point and thus $\textit{every}$ mode, with a lower-bounded generation probability. Constructing the generator mixture has a connection to the multiplicative weights update rule, upon which we propose our algorithm. We prove that our algorithm guarantees complete mode coverage. And our experiments on real and synthetic datasets confirm better mode coverage over recent approaches, ones that also use generator mixtures but rely on global statistical distances.

stat.ML↗

arXiv2019

Mechanics-Aware Modeling of Cloth Appearance

Micro-appearance models have brought unprecedented fidelity and details to cloth rendering. Yet, these models neglect fabric mechanics: when a piece of cloth interacts with the environment, its yarn and fiber arrangement usually changes in response to external contact and tension forces. Since subtle changes of a fabric's microstructures can greatly affect its macroscopic appearance, mechanics-driven appearance variation of fabrics has been a phenomenon that remains to be captured. We introduce a mechanics-aware model that adapts the microstructures of cloth yarns in a physics-based manner. Our technique works on two distinct physical scales: using physics-based simulations of individual yarns, we capture the rearrangement of yarn-level structures in response to external forces. These yarn structures are further enriched to obtain appearance-driving fiber-level details. The cross-scale enrichment is made practical through a new parameter fitting algorithm for simulation, an augmented procedural yarn model coupled with a custom-design regression neural network. We train the network using a dataset generated by joint simulations at both the yarn and the fiber levels. Through several examples, we demonstrate that our model is capable of synthesizing photorealistic cloth appearance in a %dynamic and mechanically plausible way.

cs.GR↗

arXiv2019

Disentangling the effects of doping, strain and defects in monolayer WS2 by optical spectroscopy

Monolayers of transition metal dichalcogenides (TMdC) are promising candidates for realization of a new generation of optoelectronic devices. The optical properties of these two-dimensional materials, however, vary from flake to flake, or even across individual flakes, and change over time, all of which makes control of the optoelectronic properties challenging. There are many different perturbations that can alter the optical properties, including charge doping, defects, strain, oxidation, and water intercalation. Identifying which perturbations are present is usually not straightforward and requires multiple measurements using multiple experimental modalities, which presents barriers when attempting to optimise preparation of these materials. Here, we apply highresolution photoluminescence and differential reflectance hyperspectral imaging in situ to CVD-grown WS2 monolayers. By combining these two optical measurements and using a statistical correlation analysis we are able to disentangle three contributions modulating optoelectronic properties of these materials: electron doping, strain and defects. In separating these contributions, we also observe that the B-exciton energy is less sensitive to variations in doping density than A-excitons.

cond-mat.mtrl-sci↗

arXiv2019

Dielectric Metasurfaces for Complete and Independent Control of Optical Amplitude and Phase

Metasurfaces are optically thin metamaterials that promise complete control of the wavefront of light but are primarily used to control only the phase of light. Here, we present an approach, simple in concept and in practice, that uses meta-atoms with a varying degree of form birefringence and rotation angles to create high-efficiency dielectric metasurfaces that control both the optical amplitude and phase at one or two frequencies. This opens up applications in computer-generated holography, allowing faithful reproduction of both the phase and amplitude of a target holographic scene without the iterative algorithms required in phase-only holography. We demonstrate all-dielectric metasurface holograms with independent and complete control of the amplitude and phase at up to two optical frequencies simultaneously to generate two- and three-dimensional holographic objects. We show that phase-amplitude metasurfaces enable a few features not attainable in phase-only holography; these include creating artifact-free two-dimensional holographic images, encoding phase and amplitude profiles separately at the object plane, encoding intensity profiles at the metasurface and object planes separately, and controlling the surface textures of three-dimensional holographic objects.

physics.optics↗

arXiv2018

BourGAN: Generative Networks with Metric Embeddings

This paper addresses the mode collapse for generative adversarial networks (GANs). We view modes as a geometric structure of data distribution in a metric space. Under this geometric lens, we embed subsamples of the dataset from an arbitrary metric space into the l2 space, while preserving their pairwise distance distribution. Not only does this metric embedding determine the dimensionality of the latent space automatically, it also enables us to construct a mixture of Gaussians to draw latent space random vectors. We use the Gaussian mixture model in tandem with a simple augmentation of the objective function to train GANs. Every major step of our method is supported by theoretical analysis, and our experiments on real and synthetic data confirm that the generator is able to produce samples spreading over most of the modes while avoiding unwanted samples, outperforming several recent GAN variants on a number of metrics and offering new features.

stat.ML↗

arXiv2018

Scene-Aware Audio for 360\textdegree{} Videos

Although 360\textdegree{} cameras ease the capture of panoramic footage, it remains challenging to add realistic 360\textdegree{} audio that blends into the captured scene and is synchronized with the camera motion. We present a method for adding scene-aware spatial audio to 360\textdegree{} videos in typical indoor scenes, using only a conventional mono-channel microphone and a speaker. We observe that the late reverberation of a room's impulse response is usually diffuse spatially and directionally. Exploiting this fact, we propose a method that synthesizes the directional impulse response between any source and listening locations by combining a synthesized early reverberation part and a measured late reverberation tail. The early reverberation is simulated using a geometric acoustic simulation and then enhanced using a frequency modulation method to capture room resonances. The late reverberation is extracted from a recorded impulse response, with a carefully chosen time duration that separates out the late reverberation from the early reverberation. In our validations, we show that our synthesized spatial audio matches closely with recordings using ambisonic microphones. Lastly, we demonstrate the strength of our method in several applications.

cs.GR↗

arXiv2017

AirCode: Unobtrusive Physical Tags for Digital Fabrication

We present AirCode, a technique that allows the user to tag physically fabricated objects with given information. An AirCode tag consists of a group of carefully designed air pockets placed beneath the object surface. These air pockets are easily produced during the fabrication process of the object, without any additional material or postprocessing. Meanwhile, the air pockets affect only the scattering light transport under the surface, and thus are hard to notice to our naked eyes. But, by using a computational imaging method, the tags become detectable. We present a tool that automates the design of air pockets for the user to encode information. AirCode system also allows the user to retrieve the information from captured images via a robust decoding algorithm. We demonstrate our tagging technique with applications for metadata embedding, robotic grasping, as well as conveying object affordances.

cs.HC↗

arXiv2017

FontCode: Embedding Information in Text Documents using Glyph Perturbation

We introduce FontCode, an information embedding technique for text documents. Provided a text document with specific fonts, our method embeds user-specified information in the text by perturbing the glyphs of text characters while preserving the text content. We devise an algorithm to chooses unobtrusive yet machine-recognizable glyph perturbations, leveraging a recently developed generative model that alters the glyphs of each character continuously on a font manifold. We then introduce an algorithm that embeds a user-provided message in the text document and produces an encoded document whose appearance is minimally perturbed from the original document. We also present a glyph recognition method that recovers the embedded information from an encoded document stored as a vector graphic or pixel image, or even on a printed paper. In addition, we introduce a new error-correction coding scheme that rectifies a certain number of recognition errors. Lastly, we demonstrate that our technique enables a wide array of applications, using it as a text document metadata holder, an unobtrusive optical barcode, a cryptographic message embedding scheme, and a text document signature.

cs.CV↗

arXiv2016

Room temperature single photon emission from oxidized tungsten disulphide multilayers

Two dimensional systems offer a unique platform to study light matter interaction at the nanoscale. In this work we report on robust quantum emitters fabricated by thermal oxidation of tungsten disulphide multilayers. The emitters show robust, optically stable, linearly polarized luminescence at room temperature, can be modeled using a three level system, and exhibit moderate bunching. Overall, our results provide important insights into understanding of defect formation and quantum emitter activation in 2D materials.

cond-mat.mes-hall↗

arXiv2016

Inverse Diffusion Curves using Shape Optimization

The inverse diffusion curve problem focuses on automatic creation of diffusion curve images that resemble user provided color fields. This problem is challenging since the 1D curves have a nonlinear and global impact on resulting color fields via a partial differential equation (PDE). We introduce a new approach complementary to previous methods by optimizing curve geometry. In particular, we propose a novel iterative algorithm based on the theory of shape derivatives. The resulting diffusion curves are clean and well-shaped, and the final image closely approximates the input. Our method provides a user-controlled parameter to regularize curve complexity, and generalizes to handle input color fields represented in a variety of formats.

cs.GR↗

arXiv2015

Helicity resolved Raman scattering of MoS2, MoSe2, WS2 and WSe2 atomic layers

The two-fold valley degeneracy in two-dimensional (2D) semiconducting transition metal dichalcogenides (TMDCs) (Mo,W)(S,Se)2 is suitable for "valleytronics", the storage and manupulation of information utilizing the valley degree of freedom. The conservation of luminescent photon helicity in these 2D crystal monolayers has been widely regarded as a benchmark indicator for charge carrier valley polarization. Here we perform helicity-resolved Raman scattering of the TMDC atomic layers. In drastic contrast to luminescence, the dominant first-order zone-center Raman bands, including the low energy breathing and shear modes as well as the higher energy optical phonons, are found to either maintain or completely switch the helicity of incident photons. These experimental observations, in addition to providing a useful tool for characterization of TMDC atomic layers, shed new light on the connection between photon helicity and valley polarization.

cond-mat.mes-hall↗

arXiv2013

A fast implicit method for time-dependent Hamilton-Jacobi PDEs

We present a new efficient computational approach for time-dependent first-order Hamilton-Jacobi-Bellman PDEs. Since our method is based on a time-implicit Eulerian discretization, the numerical scheme is unconditionally stable, but discretized equations for each time-slice are coupled and non-linear. We show that the same system can be re-interpreted as a discretization of a static Hamilton-Jacobi-Bellman PDE on the same physical domain. The latter was shown to be ``causal' in [Vladimirsky 2006], making fast (non-iterative)methods applicable. The implicit discretization results in higher computational cost per time slice compared to the explicit time marching. However, the latter is subject to a CFL-stability condition, and the implicit approach becomes significantly more efficient whenever the accuracy demands on the time-step are less restrictive than the stability. We also present a hybrid method, which aims to combine the advantages of both the explicit and implicit discretizations. We demonstrate the efficiency of our approach using several examples in optimal control of isotropic fixed-horizon processes.

math.NA↗

arXiv2012

Learning to Place New Objects in a Scene

Placing is a necessary skill for a personal robot to have in order to perform tasks such as arranging objects in a disorganized room. The object placements should not only be stable but also be in their semantically preferred placing areas and orientations. This is challenging because an environment can have a large variety of objects and placing areas that may not have been seen by the robot before. In this paper, we propose a learning approach for placing multiple objects in different placing areas in a scene. Given point-clouds of the objects and the scene, we design appropriate features and use a graphical model to encode various properties, such as the stacking of objects, stability, object-area relationship and common placing constraints. The inference in our model is an integer linear program, which we solve efficiently via an LP relaxation. We extensively evaluate our approach on 98 objects from 16 categories being placed into 40 areas. Our robotic experiments show a success rate of 98% in placing known objects and 82% in placing new objects stably. We use our method on our robots for performing tasks such as loading several dish-racks, a bookshelf and a fridge with multiple items.

cs.RO↗

arXiv2011

Learning to Place New Objects

The ability to place objects in the environment is an important skill for a personal robot. An object should not only be placed stably, but should also be placed in its preferred location/orientation. For instance, a plate is preferred to be inserted vertically into the slot of a dish-rack as compared to be placed horizontally in it. Unstructured environments such as homes have a large variety of object types as well as of placing areas. Therefore our algorithms should be able to handle placing new object types and new placing areas. These reasons make placing a challenging manipulation task. In this work, we propose a supervised learning algorithm for finding good placements given the point-clouds of the object and the placing area. It learns to combine the features that capture support, stability and preferred placements using a shared sparsity structure in the parameters. Even when neither the object nor the placing area is seen previously in the training set, our algorithm predicts good placements. In extensive experiments, our method enables the robot to stably place several new objects in several new placing areas with 98% success-rate; and it placed the objects in their preferred placements in 92% of the cases.

cs.RO↗