arXiv subjects

Chen Chen

Publications and source records attributed to Chen Chen.

At least 901 records · Page 50Linked to original sources

arXiv2020

A3D: Adaptive 3D Networks for Video Action Recognition

This paper presents A3D, an adaptive 3D network that can infer at a wide range of computational constraints with one-time training. Instead of training multiple models in a grid-search manner, it generates good configurations by trading off between network width and spatio-temporal resolution. Furthermore, the computation cost can be adapted after the model is deployed to meet variable constraints, for example, on edge devices. Even under the same computational constraints, the performance of our adaptive networks can be significantly boosted over the baseline counterparts by the mutual training along three dimensions. When a multiple pathway framework, e.g. SlowFast, is adopted, our adaptive method encourages a better trade-off between pathways than manual designs. Extensive experiments on the Kinetics dataset show the effectiveness of the proposed framework. The performance gain is also verified to transfer well between datasets and tasks. Code will be made available.

cs.CV↗

arXiv2020

Cross-directional Feature Fusion Network for Building Damage Assessment from Satellite Imagery

Fast and effective responses are required when a natural disaster (e.g., earthquake, hurricane, etc.) strikes. Building damage assessment from satellite imagery is critical before an effective response is conducted. High-resolution satellite images provide rich information with pre- and post-disaster scenes for analysis. However, most existing works simply use pre- and post-disaster images as input without considering their correlations. In this paper, we propose a novel cross-directional fusion strategy to better explore the correlations between pre- and post-disaster images. Moreover, the data augmentation method CutMix is exploited to tackle the challenge of hard classes. The proposed method achieves state-of-the-art performance on a large-scale building damage assessment dataset -- xBD.

cs.CV↗

arXiv2020

Demonstration of imaginary-mass particles by optical simulation in non-Hermitian systems

Unifying quantum mechanics and special relativity, the Dirac equation describes the behaviour of relativistic quantum particles, including imaginary-mass particles with faster-than-light speeds (e.g., tachyon). However, experimental searches for such particles remain negative so far. Fortunately, there are possibilities to investigate the relativistic quantum effects by quantum simulations using a controllable system. However, these simulations are mostly explored in closed physical systems that hinders the demonstration of imaginary-mass particles, whereas the general open framework would promise more possibilities in quantum simulations. Here, we reveal that the dynamic behaviors of imaginary-mass particles can be mapped to the transfer of wave packet in a non-Hermitian silicon optical lattice. We experimentally demonstrate a super-divergent dynamics of tachyon that travels faster than massless particles (e.g., photon). Besides, our simulation suggests another kind of imaginary-mass particles that have rarely been explored, which show non-divergent behaviors with imaginary energy. It is the first experimental quantum simulation being able to observe the imaginary-mass particles that have never been experimentally verified. Our work provides a fully controllable and extensible platform to investigate relativistic quantum phenomena at a chip-scale level, which would inspire more insightful explorations in quantum effects with non-Hermiticity.

physics.optics↗

arXiv2020

High efficient metasurface quarter-wave plate with wavefront engineering

Metasurfaces with local phase tuning by subwavelength elements promise unprecedented possibilities for ultra-thin and multifunctional optical devices, in which geometric phase design is widely used due to its resonant-free and large tolerance in fabrications. By arranging the orientations of anisotropic nano-antennas, the geometric phase-based metasurfaces can convert the incident spin light to its orthogonal state, and enable flexible wavefront engineering together with the function of a half-wave plate. Here, by incorporating the propagation phase, we realize another important optical device of quarter-wave plate together with the wavefront engineering as well, which is implemented by controlling both the cross- and co-polarized light simultaneously with a singlet metasurface. Highly efficient conversion of the spin light to a variety of linearly polarized light are obtained for meta-holograms, metalens focusing and imaging in blue light region. Our work provides a new strategy for efficient metasurfaces with both phase and polarization control, and enriches the functionalities of metasurface devices for wider application scenarios.

physics.optics↗

arXiv2020

Towards Resolving the Challenge of Long-tail Distribution in UAV Images for Object Detection

Existing methods for object detection in UAV images ignored an important challenge - imbalanced class distribution in UAV images - which leads to poor performance on tail classes. We systematically investigate existing solutions to long-tail problems and unveil that re-balancing methods that are effective on natural image datasets cannot be trivially applied to UAV datasets. To this end, we rethink long-tailed object detection in UAV images and propose the Dual Sampler and Head detection Network (DSHNet), which is the first work that aims to resolve long-tail distribution in UAV images. The key components in DSHNet include Class-Biased Samplers (CBS) and Bilateral Box Heads (BBH), which are developed to cope with tail classes and head classes in a dual-path manner. Without bells and whistles, DSHNet significantly boosts the performance of tail classes on different detection frameworks. Moreover, DSHNet significantly outperforms base detectors and generic approaches for long-tail problems on VisDrone and UAVDT datasets. It achieves new state-of-the-art performance when combining with image cropping methods. Code is available at https://github.com/we1pingyu/DSHNet

cs.CV↗

arXiv2020

A difference-of-convex approach for split feasibility with applications to matrix factorizations and outlier detection

The split feasibility problem is to find an element in the intersection of a closed set $C$ and the linear preimage of another closed set $D$, assuming the projections onto $C$ and $D$ are easy to compute. This class of problems arises naturally in many contemporary applications such as compressed sensing. While the sets $C$ and $D$ are typically assumed to be convex in the literature, in this paper, we allow both sets to be possibly nonconvex. We observe that, in this setting, the split feasibility problem can be formulated as an optimization problem with a difference-of-convex objective so that standard majorization-minimization type algorithms can be applied. Here we focus on the nonmonotone proximal gradient algorithm with majorization studied in [15, Appendix A]. We show that, when this algorithm is applied to a split feasibility problem, the sequence generated clusters at a stationary point of the problem under mild assumptions. We also study local convergence property of the sequence under suitable assumptions on the closed sets involved. Finally, we perform numerical experiments to illustrate the efficiency of our approach on solving split feasibility problems that arise in completely positive matrix factorization, (uniformly) sparse matrix factorization, and outlier detection.

math.OC↗

arXiv2020

Stage Lookup: Accelerating Path Lookup using Directory Shortcuts

The lookup procedure in Linux costs a significant portion of file accessing time as the virtual file system (VFS) traverses the file path components one after another. The lookup procedure becomes more time consuming when applications frequently access files, especially those with small sizes. We propose Stage Lookup, which dynamically caches popular directories to speed up lookup procedures and further reduce file accessing latency. The core of Stage Lookup is to cache popular dentries as shortcuts, so that path walks do not bother to traverse directory trees from the root. Furthermore, Stage Lookup enriches backward path walks as it treats the directory tree in a VFS as an undirected map. We implement a Stage Lookup prototype and integrate it into Linux Kernel v3.14. Our extensive performance evaluation studies show that Stage Lookup offers up to 46.9% performance gain compared to ordinary path lookup schemes. Furthermore, Stage Lookup shows smaller performance overheads in rename and chmod operations compared to the original method of the kernel.

cs.OS↗

arXiv2020

Absorption-line Environments of High-redshift BOSS Quasars

The early stage of massive galaxy evolution often involves outflows driven by a starburst or a central quasar plus cold mode accretion (infall), which adds to the mass build-up in the galaxies. To study the nature of these infall and outflows in the quasar environments, we have examined the correlation of narrow absorption lines (NALs) at positive and negative velocity shifts to other quasar properties, such as their broad absorption-line (BAL) outflows and radio-loudness, using spectral data from SDSS-BOSS DR12. Our results show that the incidence of associated absorption lines (AALs) and outflow AALs is strongly correlated with BALs, which indicates most AALs form in quasar-driven outflows. Multiple AALs are also strongly correlated with BALs, demonstrating quasar outflows tend to be highly structured and can create multiple gas components with different velocity shifts along our line of sight. Infall AALs appear less often in quasars with BALs than quasars without BALs. This suggests that BAL outflows act on large scale in host galaxies and inhibit the infall of gas from the IGM, supporting theoretical models in which quasar outflow plays an important role in the feedback to host galaxies. Despite having larger distances, infall AALs are more highly ionized than outflow AALs, which can be attributed to the lower densities in the infall absorbers.

astro-ph.GA↗

arXiv2020

GradAug: A New Regularization Method for Deep Neural Networks

We propose a new regularization method to alleviate over-fitting in deep neural networks. The key idea is utilizing randomly transformed training samples to regularize a set of sub-networks, which are originated by sampling the width of the original network, in the training process. As such, the proposed method introduces self-guided disturbances to the raw gradients of the network and therefore is termed as Gradient Augmentation (GradAug). We demonstrate that GradAug can help the network learn well-generalized and more diverse representations. Moreover, it is easy to implement and can be applied to various structures and applications. GradAug improves ResNet-50 to 78.79% on ImageNet classification, which is a new state-of-the-art accuracy. By combining with CutMix, it further boosts the performance to 79.67%, which outperforms an ensemble of advanced training tricks. The generalization ability is evaluated on COCO object detection and instance segmentation where GradAug significantly surpasses other state-of-the-art methods. GradAug is also robust to image distortions and FGSM adversarial attacks and is highly effective in low data regimes. Code is available at https://github.com/taoyang1122/GradAug

cs.CV↗

arXiv2020

Gauge-induced Floquet topological states in photonic waveguides

Tremendous efforts have been devoted to the search for exotic topological states, which usually exist at an interface between lattices with differing topological invariants according to the bulk-edge correspondence. Here, we show a new finding of topological states localized at the interface between two gauge-shifted Floquet photonic lattices, despite the same topological order across the entire structure. The quasienergy band structures reveal that these new interface modes belong to the Floquet π modes, which are further found to enable a robust one-way propagation thanks to the flexible control of the Floquet gauge. The intriguing propagations of these π-interface modes are experimentally verified in a silicon waveguides platform at near-infrared wavelengths, which show both broad working bandwidth and high tolerance to the structural fluctuations. Our approach provides a new route for light manipulations with robust behaviours in Floquet engineering and beyond.

physics.optics↗

arXiv2020

Adversarial Self-Supervised Data-Free Distillation for Text Classification

Large pre-trained transformer-based language models have achieved impressive results on a wide range of NLP tasks. In the past few years, Knowledge Distillation(KD) has become a popular paradigm to compress a computationally expensive model to a resource-efficient lightweight model. However, most KD algorithms, especially in NLP, rely on the accessibility of the original training dataset, which may be unavailable due to privacy issues. To tackle this problem, we propose a novel two-stage data-free distillation method, named Adversarial self-Supervised Data-Free Distillation (AS-DFD), which is designed for compressing large-scale transformer-based models (e.g., BERT). To avoid text generation in discrete space, we introduce a Plug & Play Embedding Guessing method to craft pseudo embeddings from the teacher's hidden knowledge. Meanwhile, with a self-supervised module to quantify the student's ability, we adapt the difficulty of pseudo embeddings in an adversarial training manner. To the best of our knowledge, our framework is the first data-free distillation framework designed for NLP tasks. We verify the effectiveness of our method on several text classification datasets.

cs.CL↗

arXiv2020

Self-Supervision with Superpixels: Training Few-shot Medical Image Segmentation without Annotation

Few-shot semantic segmentation (FSS) has great potential for medical imaging applications. Most of the existing FSS techniques require abundant annotated semantic classes for training. However, these methods may not be applicable for medical images due to the lack of annotations. To address this problem we make several contributions: (1) A novel self-supervised FSS framework for medical images in order to eliminate the requirement for annotations during training. Additionally, superpixel-based pseudo-labels are generated to provide supervision; (2) An adaptive local prototype pooling module plugged into prototypical networks, to solve the common challenging foreground-background imbalance problem in medical image segmentation; (3) We demonstrate the general applicability of the proposed approach for medical images using three different tasks: abdominal organ segmentation for CT and MRI, as well as cardiac segmentation for MRI. Our results show that, for medical image segmentation, the proposed method outperforms conventional FSS methods which require manual annotations for training.

cs.CV↗

arXiv2020

Towards Chirality Control of Graphene Nanoribbons Embedded in Hexagonal Boron Nitride

The integrated inplane growth of two dimensional materials with similar lattices, but distinct electrical properties, could provide a promising route to achieve integrated circuitry of atomic thickness. However, fabrication of edge specific GNR in the lattice of hBN still remains an enormous challenge for present approaches. Here we developed a two step growth method and successfully achieved sub 5 nm wide zigzag and armchair GNRs embedded in hBN, respectively. Further transport measurements reveal that the sub 7 nm wide zigzag GNRs exhibit openings of the band gap inversely proportional to their width, while narrow armchair GNRs exhibit some fluctuation in the bandgap width relationship.This integrated lateral growth of edge specific GNRs in hBN brings semiconducting building blocks to atomically thin layer, and will provide a promising route to achieve intricate nanoscale electrical circuits on high quality insulating hBN substrates.

cond-mat.mes-hall↗

arXiv2020

Progressive Bilateral-Context Driven Model for Post-Processing Person Re-Identification

Most existing person re-identification methods compute pairwise similarity by extracting robust visual features and learning the discriminative metric. Owing to visual ambiguities, these content-based methods that determine the pairwise relationship only based on the similarity between them, inevitably produce a suboptimal ranking list. Instead, the pairwise similarity can be estimated more accurately along the geodesic path of the underlying data manifold by exploring the rich contextual information of the sample. In this paper, we propose a lightweight post-processing person re-identification method in which the pairwise measure is determined by the relationship between the sample and the counterpart's context in an unsupervised way. We translate the point-to-point comparison into the bilateral point-to-set comparison. The sample's context is composed of its neighbor samples with two different definition ways: the first order context and the second order context, which are used to compute the pairwise similarity in sequence, resulting in a progressive post-processing model. The experiments on four large-scale person re-identification benchmark datasets indicate that (1) the proposed method can consistently achieve higher accuracies by serving as a post-processing procedure after the content-based person re-identification methods, showing its state-of-the-art results, (2) the proposed lightweight method only needs about 6 milliseconds for optimizing the ranking results of one sample, showing its high-efficiency. Code is available at: https://github.com/123ci/PBCmodel.

cs.CV↗

arXiv2020

PEARL: Plausibly Deniable Flash Translation Layer using WOM coding

When adversaries are powerful enough to coerce users to reveal encryption keys, encryption alone becomes insufficient for data protection. Plausible deniability (PD) mechanisms resolve this by enabling users to hide the mere existence of sensitive data, often by providing plausible "cover texts" or "public data volumes" hosted on the same device. Unfortunately, with the increasing prevalence of (NAND) flash as a high-performance cost-effective storage medium, PD becomes even more challenging in the presence of realistic adversaries who can usually access a device at multiple points in time ("multi-snapshot"). This is because read/write operations to flash do not result in intuitive corresponding changes to the underlying device state. The problem is further compounded by the fact that this behavior is mostly proprietary. For example, in a majority of commercially-available flash devices, an issued delete or overwrite operation from the upper layers almost certainly won't result in an actual immediate erase of the underlying flash cells. To address these challenges, we designed a new class of write-once memory (WOM) codes to store hidden bits in the same physical locations as other public bits. This is made possible by the inherent nature of NAND flash and the possibility of issuing multiple writes to target cells that have not previous been written to in existing pages. We designed PEARL, a general-purpose Flash Translation Layer (FTL) that allows users to plausibly deniably store hidden data in NAND flash devices. We implemented and evaluated PEARL on a widely used simulator FlashSim (Kim et al. 2019). PEARL performs well on real-world workloads, comparably to non-PD baselines. PEARL is the first system that achieves strong plausible deniability for NAND flash devices, secure against realistic multi-snapshot adversaries.

cs.CR↗

arXiv2020

PCA-SRGAN: Incremental Orthogonal Projection Discrimination for Face Super-resolution

Generative Adversarial Networks (GAN) have been employed for face super resolution but they bring distorted facial details easily and still have weakness on recovering realistic texture. To further improve the performance of GAN based models on super-resolving face images, we propose PCA-SRGAN which pays attention to the cumulative discrimination in the orthogonal projection space spanned by PCA projection matrix of face data. By feeding the principal component projections ranging from structure to details into the discriminator, the discrimination difficulty will be greatly alleviated and the generator can be enhanced to reconstruct clearer contour and finer texture, helpful to achieve the high perception and low distortion eventually. This incremental orthogonal projection discrimination has ensured a precise optimization procedure from coarse to fine and avoids the dependence on the perceptual regularization. We conduct experiments on CelebA and FFHQ face datasets. The qualitative visual effect and quantitative evaluation have demonstrated the overwhelming performance of our model over related works.

cs.CV↗

arXiv2020

Stochastic Adaptive Line Search for Differentially Private Optimization

The performance of private gradient-based optimization algorithms is highly dependent on the choice of step size (or learning rate) which often requires non-trivial amount of tuning. In this paper, we introduce a stochastic variant of classic backtracking line search algorithm that satisfies Rényi differential privacy. Specifically, the proposed algorithm adaptively chooses the step size satsisfying the the Armijo condition (with high probability) using noisy gradients and function estimates. Furthermore, to improve the probability with which the chosen step size satisfies the condition, it adjusts per-iteration privacy budget during runtime according to the reliability of noisy gradient. A naive implementation of the backtracking search algorithm may end up using unacceptably large privacy budget as the ability of adaptive step size selection comes at the cost of extra function evaluations. The proposed algorithm avoids this problem by using the sparse vector technique combined with the recent privacy amplification lemma. We also introduce a privacy budget adaptation strategy in which the algorithm adaptively increases the budget when it detects that directions pointed by consecutive gradients are drastically different. Extensive experiments on both convex and non-convex problems show that the adaptively chosen step sizes allow the proposed algorithm to efficiently use the privacy budget and show competitive performance against existing private optimizers.

cs.LG↗

arXiv2020

Go Wide, Then Narrow: Efficient Training of Deep Thin Networks

For deploying a deep learning model into production, it needs to be both accurate and compact to meet the latency and memory constraints. This usually results in a network that is deep (to ensure performance) and yet thin (to improve computational efficiency). In this paper, we propose an efficient method to train a deep thin network with a theoretic guarantee. Our method is motivated by model compression. It consists of three stages. First, we sufficiently widen the deep thin network and train it until convergence. Then, we use this well-trained deep wide network to warm up (or initialize) the original deep thin network. This is achieved by layerwise imitation, that is, forcing the thin network to mimic the intermediate outputs of the wide network from layer to layer. Finally, we further fine tune this already well-initialized deep thin network. The theoretical guarantee is established by using the neural mean field analysis. It demonstrates the advantage of our layerwise imitation approach over backpropagation. We also conduct large-scale empirical experiments to validate the proposed method. By training with our method, ResNet50 can outperform ResNet101, and BERT Base can be comparable with BERT Large, when ResNet101 and BERT Large are trained under the standard training procedures as in the literature.

cs.LG↗