arXiv subjects

Yang Song

Publications and source records attributed to Yang Song.

At least 343 records · Page 19Linked to original sources

arXiv2019

Class-Balanced Loss Based on Effective Number of Samples

With the rapid increase of large-scale, real-world datasets, it becomes critical to address the problem of long-tailed data distribution (i.e., a few classes account for most of the data, while most classes are under-represented). Existing solutions typically adopt class re-balancing strategies such as re-sampling and re-weighting based on the number of observations for each class. In this work, we argue that as the number of samples increases, the additional benefit of a newly added data point will diminish. We introduce a novel theoretical framework to measure data overlap by associating with each sample a small neighboring region rather than a single point. The effective number of samples is defined as the volume of samples and can be calculated by a simple formula $(1-β^{n})/(1-β)$, where $n$ is the number of samples and $β\in [0,1)$ is a hyperparameter. We design a re-weighting scheme that uses the effective number of samples for each class to re-balance the loss, thereby yielding a class-balanced loss. Comprehensive experiments are conducted on artificially induced long-tailed CIFAR datasets and large-scale datasets including ImageNet and iNaturalist. Our results show that when trained with the proposed class-balanced loss, the network is able to achieve significant performance gains on long-tailed datasets.

cs.CV↗

arXiv2018

Description of Stability for Linear Time-Invariant Systems Based on the First Curvature

This paper focuses on using the first curvature $κ(t)$ of trajectory to describe the stability of linear time-invariant system. We extend the results for two and three-dimensional systems [Y. Wang, H. Sun, Y. Song et al., arXiv:1808.00290] to $n$-dimensional systems. We prove that for a system $\dot{r}(t)=Ar(t)$, (i) if there exists a measurable set whose Lebesgue measure is greater than zero, such that for all initial values in this set, $\lim\limits_{t\to+\infty}κ(t)\neq0$ or $\lim\limits_{t\to+\infty}κ(t)$ does not exist, then the zero solution of the system is stable; (ii) if the matrix $A$ is invertible, and there exists a measurable set whose Lebesgue measure is greater than zero, such that for all initial values in this set, $\lim\limits_{t\to+\infty}κ(t)=+\infty$, then the zero solution of the system is asymptotically stable.

math.OC↗

arXiv2018

Beyond Inferring Class Representatives: User-Level Privacy Leakage From Federated Learning

Federated learning, i.e., a mobile edge computing framework for deep learning, is a recent advance in privacy-preserving machine learning, where the model is trained in a decentralized manner by the clients, i.e., data curators, preventing the server from directly accessing those private data from the clients. This learning mechanism significantly challenges the attack from the server side. Although the state-of-the-art attacking techniques that incorporated the advance of Generative adversarial networks (GANs) could construct class representatives of the global data distribution among all clients, it is still challenging to distinguishably attack a specific client (i.e., user-level privacy leakage), which is a stronger privacy threat to precisely recover the private data from a specific client. This paper gives the first attempt to explore user-level privacy leakage against the federated learning by the attack from a malicious server. We propose a framework incorporating GAN with a multi-task discriminator, which simultaneously discriminates category, reality, and client identity of input samples. The novel discrimination on client identity enables the generator to recover user specified private data. Unlike existing works that tend to interfere the training process of the federated learning, the proposed method works "invisibly" on the server side. The experimental results demonstrate the effectiveness of the proposed attacking approach and the superior to the state-of-the-art.

cs.LG↗

arXiv2018

Constructing Unrestricted Adversarial Examples with Generative Models

Adversarial examples are typically constructed by perturbing an existing data point within a small matrix norm, and current defense methods are focused on guarding against this type of attack. In this paper, we propose unrestricted adversarial examples, a new threat model where the attackers are not restricted to small norm-bounded perturbations. Different from perturbation-based attacks, we propose to synthesize unrestricted adversarial examples entirely from scratch using conditional generative models. Specifically, we first train an Auxiliary Classifier Generative Adversarial Network (AC-GAN) to model the class-conditional distribution over data samples. Then, conditioned on a desired class, we search over the AC-GAN latent space to find images that are likely under the generative model and are misclassified by a target classifier. We demonstrate through human evaluation that unrestricted adversarial examples generated this way are legitimate and belong to the desired class. Our empirical results on the MNIST, SVHN, and CelebA datasets show that unrestricted adversarial examples can bypass strong adversarial training and certified defense methods designed for traditional adversarial attacks.

cs.LG↗

arXiv2018

UWB/LiDAR Fusion For Cooperative Range-Only SLAM

We equip an ultra-wideband (UWB) node and a 2D LiDAR sensor a.k.a. 2D laser rangefinder on a mobile robot, and place UWB beacon nodes at unknown locations in an unknown environment. All UWB nodes can do ranging with each other thus forming a cooperative sensor network. We propose to fuse the peer-to-peer ranges measured between UWB nodes and laser scanning information, i.e. range measured between robot and nearby objects/obstacles, for simultaneous localization of the robot, all UWB beacons, and LiDAR mapping. The fusion is inspired by two facts: 1) LiDAR may improve UWB-only localization accuracy as it gives a more precise and comprehensive picture of the surrounding environment; 2) on the other hand, UWB ranging measurements may remove the error accumulated in the LiDAR-based SLAM algorithm. Our experiments demonstrate that UWB/LiDAR fusion enables drift-free SLAM in real-time based on ranging measurements only.

cs.RO↗

arXiv2018

Metallic interfaces in a CaTiO$_3$/LaTiO$_3$ heterostructure

Almost all oxide two-dimensional electron gases are formed in SrTiO$_3$-based heterostructures and the study of non-SrTiO$_3$ systems is extremely rare. Here, we report the realization of a two-dimensional electron gas in a CaTiO$_3$-based heterostructure, CaTiO$_3$/LaTiO$_3$, grown epitaxially layer-by-layer on a NdGaO$_3$ (110) substrate via pulsed laser deposition. The high quality of the crystal and electronic structures are characterized by in-situ reflection high-energy electron diffraction, X-ray diffraction, and X-ray photoemission spectroscopy. Measurement of electrical transport validates the formation of a two-dimensional electron gas in the CaTiO$_3$/LaTiO$_3$ superlattice. It is revealed the room-temperature carrier mobility in CaTiO$_3$/LaTiO$_3$ is nearly 3 times higher than in CaTiO$_3$/YTiO$_3$, demonstrating the effect of TiO$_6$ octahedral tilts and rotations on carrier mobility of two-dimensional electron gases. Due to doped CaTiO$_3$ being an A-site polar metal, our results provide a new route to design novel A-site two-dimensional polar metals.

cond-mat.mtrl-sci↗

arXiv2018

Description of Stability for Two and Three-Dimensional Linear Time-Invariant Systems Based on Curvature and Torsion

This paper focuses on using curvature and torsion to describe the stability of linear time-invariant system. We prove that for a two-dimensional system $\dot{r}(t)= Ar(t)$, (i) if there exists an initial value, such that zero is not the limit of curvature of trajectory as $t\to+\infty$, then the zero solution of the system is stable; (ii) if there exists an initial value, such that the limit of curvature of trajectory is infinity as $t\to+\infty$, then the zero solution of the system is asymptotically stable. For a three-dimensional system, (i) if there exists a measurable set whose Lebesgue measure is greater than zero, such that for all initial values in this set, zero is not the limit of curvature of trajectory as $t\to+\infty$, then the zero solution of the system is stable; (ii) if the coefficient matrix is invertible, and there exists a measurable set whose Lebesgue measure is greater than zero, such that for all initial values in this set, the limit of curvature of trajectory is infinity as $t\to+\infty$, then the zero solution of the system is asymptotically stable; (iii) if there exists a measurable set whose Lebesgue measure is greater than zero, such that for all initial values in this set, zero is not the limit of torsion of trajectory as $t\to+\infty$, then the zero solution of the system is asymptotically stable.

math.OC↗

arXiv2018

Phonon-induced Majorana Qubit Relaxation in Tunnel-Coupled Two-Island Topological Superconductors

A Majorana-based qubit can form within the fermion parity subspace of a two-island topological superconductor setup consisting of four Majorana zero modes. We study the relaxation time of this Majorana qubit induced by the intrinsic Majorana-phonon interaction in combination with single electron tunneling between the two islands. The theory is developed for both the spinless p-wave Kitaev chain model and the spin-orbit-coupled semiconductor nanowire with induced superconductivity and Zeeman splitting. We systematically discuss the dependence of the relaxation rate on Majorana charge distribution (varying with the chemical potential, magnetic field, superconducting paring, spin-orbit coupling and island length) as well as on phonon wavelength and the length of the tunnel barrier. Importantly, the accompanied phonon energy can be tuned by the voltage bias over the tunnel barrier, leading to relatively easy manipulation and suppression of the relaxation rate.

cond-mat.mes-hall↗

arXiv2018

Deep Dual Pyramid Network for Barcode Segmentation using Barcode-30k Database

Digital signs(such as barcode or QR code) are widely used in our daily life, and for many applications, we need to localize them on images. However, difficult cases such as targets with small scales, half-occlusion, shape deformation and large illumination changes cause challenges for conventional methods. In this paper, we address this problem by producing a large-scale dataset and adopting a deep learning based semantic segmentation approach. Specifically, a synthesizing method was proposed to generate well-annotated images containing barcode and QR code labels, which contributes to largely decrease the annotation time. Through the synthesis strategy, we introduce a dataset that contains 30000 images with Barcode and QR code - Barcode-30k. Moreover, we further propose a dual pyramid structure based segmentation network - BarcodeNet, which is mainly formed with two novel modules, Prior Pyramid Pooling Module(P3M) and Pyramid Refine Module(PRM). We validate the effectiveness of BarcodeNet on the proposed synthetic dataset, and it yields the result of mIoU accuracy 95.36\% on validation set. Additional segmentation results of real images have shown that accurate segmentation performance is achieved.

cs.CV↗

arXiv2018

3D Global Convolutional Adversarial Network\\ for Prostate MR Volume Segmentation

Advanced deep learning methods have been developed to conduct prostate MR volume segmentation in either a 2D or 3D fully convolutional manner. However, 2D methods tend to have limited segmentation performance, since large amounts of spatial information of prostate volumes are discarded during the slice-by-slice segmentation process; and 3D methods also have room for improvement, since they use isotropic kernels to perform 3D convolutions whereas most prostate MR volumes have anisotropic spatial resolution. Besides, the fully convolutional structural methods achieve good performance for localization issues but neglect the per-voxel classification for segmentation tasks. In this paper, we propose a 3D Global Convolutional Adversarial Network (3D GCA-Net) to address efficient prostate MR volume segmentation. We first design a 3D ResNet encoder to extract 3D features from prostate scans, and then develop the decoder, which is composed of a multi-scale 3D global convolutional block and a 3D boundary refinement block, to address the classification and localization issues simultaneously for volumetric segmentation. Additionally, we combine the encoder-decoder segmentation network with an adversarial network in the training phrase to enforce the contiguity of long-range spatial predictions. Throughout the proposed model, we use anisotropic convolutional processing for better feature learning on prostate MR scans. We evaluated our 3D GCA-Net model on two public prostate MR datasets and achieved state-of-the-art performances.

cs.CV↗

arXiv2018

Large Scale Fine-Grained Categorization and Domain-Specific Transfer Learning

Transferring the knowledge learned from large scale datasets (e.g., ImageNet) via fine-tuning offers an effective solution for domain-specific fine-grained visual categorization (FGVC) tasks (e.g., recognizing bird species or car make and model). In such scenarios, data annotation often calls for specialized domain knowledge and thus is difficult to scale. In this work, we first tackle a problem in large scale FGVC. Our method won first place in iNaturalist 2017 large scale species classification challenge. Central to the success of our approach is a training scheme that uses higher image resolution and deals with the long-tailed distribution of training data. Next, we study transfer learning via fine-tuning from large scale datasets to small scale, domain-specific FGVC datasets. We propose a measure to estimate domain similarity via Earth Mover's Distance and demonstrate that transfer learning benefits from pre-training on a source domain that is similar to the target domain by this measure. Our proposed transfer learning outperforms ImageNet pre-training and obtains state-of-the-art results on multiple commonly used FGVC datasets.

cs.CV↗

arXiv2018

Accelerating Natural Gradient with Higher-Order Invariance

An appealing property of the natural gradient is that it is invariant to arbitrary differentiable reparameterizations of the model. However, this invariance property requires infinitesimal steps and is lost in practical implementations with small but finite step sizes. In this paper, we study invariance properties from a combined perspective of Riemannian geometry and numerical differential equation solving. We define the order of invariance of a numerical method to be its convergence order to an invariant solution. We propose to use higher-order integrators and geodesic corrections to obtain more invariant optimization trajectories. We prove the numerical convergence properties of geodesic corrected updates and show that they can be as computationally efficient as plain natural gradient. Experimentally, we demonstrate that invariance leads to faster optimization and our techniques improve on traditional natural gradient in deep neural network training and natural policy gradient for reinforcement learning.

cs.LG↗

arXiv2018

PixelDefend: Leveraging Generative Models to Understand and Defend against Adversarial Examples

Adversarial perturbations of normal images are usually imperceptible to humans, but they can seriously confuse state-of-the-art machine learning models. What makes them so special in the eyes of image classifiers? In this paper, we show empirically that adversarial examples mainly lie in the low probability regions of the training distribution, regardless of attack types and targeted models. Using statistical hypothesis testing, we find that modern neural density models are surprisingly good at detecting imperceptible image perturbations. Based on this discovery, we devised PixelDefend, a new approach that purifies a maliciously perturbed image by moving it back towards the distribution seen in the training data. The purified image is then run through an unmodified classifier, making our method agnostic to both the classifier and the attacking method. As a result, PixelDefend can be used to protect already deployed models and be combined with other model-specific defenses. Experiments show that our method greatly improves resilience across a wide variety of state-of-the-art attacking methods, increasing accuracy on the strongest attack from 63% to 84% for Fashion MNIST and from 32% to 70% for CIFAR-10.

cs.LG↗

arXiv2018

The iNaturalist Species Classification and Detection Dataset

Existing image classification datasets used in computer vision tend to have a uniform distribution of images across object categories. In contrast, the natural world is heavily imbalanced, as some species are more abundant and easier to photograph than others. To encourage further progress in challenging real world conditions we present the iNaturalist species classification and detection dataset, consisting of 859,000 images from over 5,000 different species of plants and animals. It features visually similar species, captured in a wide variety of situations, from all over the world. Images were collected with different camera types, have varying image quality, feature a large class imbalance, and have been verified by multiple citizen scientists. We discuss the collection of the dataset and present extensive baseline experiments using state-of-the-art computer vision classification and detection models. Results show that current non-ensemble based methods achieve only 67% top one classification accuracy, illustrating the difficulty of the dataset. Specifically, we observe poor results for classes with small numbers of training examples suggesting more attention is needed in low-shot learning.

cs.CV↗

arXiv2018

SARA: Self-Aware Resource Allocation for Heterogeneous MPSoCs

In modern heterogeneous MPSoCs, the management of shared memory resources is crucial in delivering end-to-end QoS. Previous frameworks have either focused on singular QoS targets or the allocation of partitionable resources among CPU applications at relatively slow timescales. However, heterogeneous MPSoCs typically require instant response from the memory system where most resources cannot be partitioned. Moreover, the health of different cores in a heterogeneous MPSoC is often measured by diverse performance objectives. In this work, we propose a Self-Aware Resource Allocation (SARA) framework for heterogeneous MPSoCs. Priority-based adaptation allows cores to use different target performance and self-monitor their own intrinsic health. In response, the system allocates non-partitionable resources based on priorities. The proposed framework meets a diverse range of QoS demands from heterogeneous cores.

cs.DC↗

arXiv2018

Decoupled Learning for Conditional Adversarial Networks

Incorporating encoding-decoding nets with adversarial nets has been widely adopted in image generation tasks. We observe that the state-of-the-art achievements were obtained by carefully balancing the reconstruction loss and adversarial loss, and such balance shifts with different network structures, datasets, and training strategies. Empirical studies have demonstrated that an inappropriate weight between the two losses may cause instability, and it is tricky to search for the optimal setting, especially when lacking prior knowledge on the data and network. This paper gives the first attempt to relax the need of manual balancing by proposing the concept of \textit{decoupled learning}, where a novel network structure is designed that explicitly disentangles the backpropagation paths of the two losses. Experimental results demonstrate the effectiveness, robustness, and generality of the proposed method. The other contribution of the paper is the design of a new evaluation metric to measure the image quality of generative models. We propose the so-called \textit{normalized relative discriminative score} (NRDS), which introduces the idea of relative comparison, rather than providing absolute estimates like existing metrics.

cs.CV↗

arXiv2017

r-BTN: Cross-domain Face Composite and Synthesis from Limited Facial Patches

We start by asking an interesting yet challenging question, "If an eyewitness can only recall the eye features of the suspect, such that the forensic artist can only produce a sketch of the eyes (e.g., the top-left sketch shown in Fig. 1), can advanced computer vision techniques help generate the whole face image?" A more generalized question is that if a large proportion (e.g., more than 50%) of the face/sketch is missing, can a realistic whole face sketch/image still be estimated. Existing face completion and generation methods either do not conduct domain transfer learning or can not handle large missing area. For example, the inpainting approach tends to blur the generated region when the missing area is large (i.e., more than 50%). In this paper, we exploit the potential of deep learning networks in filling large missing region (e.g., as high as 95% missing) and generating realistic faces with high-fidelity in cross domains. We propose the recursive generation by bidirectional transformation networks (r-BTN) that recursively generates a whole face/sketch from a small sketch/face patch. The large missing area and the cross domain challenge make it difficult to generate satisfactory results using a unidirectional cross-domain learning structure. On the other hand, a forward and backward bidirectional learning between the face and sketch domains would enable recursive estimation of the missing region in an incremental manner (Fig. 1) and yield appealing results. r-BTN also adopts an adversarial constraint to encourage the generation of realistic faces/sketches. Extensive experiments have been conducted to demonstrate the superior performance from r-BTN as compared to existing potential solutions.

cs.CV↗

arXiv2017

Spin-orbit coupling induced two-electron relaxation in silicon donor pairs

We unravel theoretically a key intrinsic relaxation mechanism among the low-lying singlet and triplet donor-pair states in silicon, an important element in the fast-developing field of spintronics and quantum computation. Despite the perceived weak spin-orbit coupling (SOC) in Si, we find that our discovered relaxation mechanism, combined with the electron-phonon and inter-donor interactions, dominantly drives the transitions in the two-electron states over a large range of donor coupling regime. The scaling of the relaxation rate with inter-donor exchange interaction $J$ goes from $J^5$ to $J^4$ at the low to high temperature limits. Our analytical study draws on the symmetry analysis over combined band, donor envelope and valley configurations. It uncovers naturally the dependence on the donor-alignment direction and triplet spin orientation, and especially on the dominant SOC source from donor impurities. While a magnetic field is not necessary for this relaxation, unlike in the single-donor spin relaxation, we discuss the crossover behavior with increasing Zeeman energy in order to facilitate comparison with experiments.

cond-mat.mes-hall↗