arXiv subjects

Dong Xu

Publications and source records attributed to Dong Xu.

At least 199 records · Page 11Linked to original sources

arXiv2020

Simulating multi-exit evacuation using deep reinforcement learning

Conventional simulations on multi-exit indoor evacuation focus primarily on how to determine a reasonable exit based on numerous factors in a changing environment. Results commonly include some congested and other under-utilized exits, especially with massive pedestrians. We propose a multi-exit evacuation simulation based on Deep Reinforcement Learning (DRL), referred to as the MultiExit-DRL, which involves in a Deep Neural Network (DNN) framework to facilitate state-to-action mapping. The DNN framework applies Rainbow Deep Q-Network (DQN), a DRL algorithm that integrates several advanced DQN methods, to improve data utilization and algorithm stability, and further divides the action space into eight isometric directions for possible pedestrian choices. We compare MultiExit-DRL with two conventional multi-exit evacuation simulation models in three separate scenarios: 1) varying pedestrian distribution ratios, 2) varying exit width ratios, and 3) varying open schedules for an exit. The results show that MultiExit-DRL presents great learning efficiency while reducing the total number of evacuation frames in all designed experiments. In addition, the integration of DRL allows pedestrians to explore other potential exits and helps determine optimal directions, leading to the high efficiency of exit utilization.

cs.LG↗

arXiv2020

A Unified End-to-End Framework for Efficient Deep Image Compression

Image compression is a widely used technique to reduce the spatial redundancy in images. Recently, learning based image compression has achieved significant progress by using the powerful representation ability from neural networks. However, the current state-of-the-art learning based image compression methods suffer from the huge computational cost, which limits their capacity for practical applications. In this paper, we propose a unified framework called Efficient Deep Image Compression (EDIC) based on three new technologies, including a channel attention module, a Gaussian mixture model and a decoder-side enhancement module. Specifically, we design an auto-encoder style network for learning based image compression. To improve the coding efficiency, we exploit the channel relationship between latent representations by using the channel attention module. Besides, the Gaussian mixture model is introduced for the entropy model and improves the accuracy for bitrate estimation. Furthermore, we introduce the decoder-side enhancement module to further improve image compression performance. Our EDIC method can also be readily incorporated with the Deep Video Compression (DVC) framework to further improve the video compression performance. Simultaneously, our EDIC method boosts the coding performance significantly while bringing slightly increased computational cost. More importantly, experimental results demonstrate that the proposed approach outperforms the current state-of-the-art image compression methods and is up to more than 150 times faster in terms of decoding speed when compared with Minnen's method. The proposed framework also successfully improves the performance of the recent deep video compression system DVC. Our code will be released at https://github.com/liujiaheng/compression.

eess.IV↗

arXiv2020

Content Adaptive and Error Propagation Aware Deep Video Compression

Recently, learning based video compression methods attract increasing attention. However, the previous works suffer from error propagation due to the accumulation of reconstructed error in inter predictive coding. Meanwhile, the previous learning based video codecs are also not adaptive to different video contents. To address these two problems, we propose a content adaptive and error propagation aware video compression system. Specifically, our method employs a joint training strategy by considering the compression performance of multiple consecutive frames instead of a single frame. Based on the learned long-term temporal information, our approach effectively alleviates error propagation in reconstructed frames. More importantly, instead of using the hand-crafted coding modes in the traditional compression systems, we design an online encoder updating scheme in our system. The proposed approach updates the parameters for encoder according to the rate-distortion criterion but keeps the decoder unchanged in the inference stage. Therefore, the encoder is adaptive to different video contents and achieves better compression performance by reducing the domain gap between the training and testing datasets. Our method is simple yet effective and outperforms the state-of-the-art learning based video codecs on benchmark datasets without increasing the model size or decreasing the decoding speed.

eess.IV↗

arXiv2020

Channel Pruning Guided by Classification Loss and Feature Importance

In this work, we propose a new layer-by-layer channel pruning method called Channel Pruning guided by classification Loss and feature Importance (CPLI). In contrast to the existing layer-by-layer channel pruning approaches that only consider how to reconstruct the features from the next layer, our approach additionally take the classification loss into account in the channel pruning process. We also observe that some reconstructed features will be removed at the next pruning stage. So it is unnecessary to reconstruct these features. To this end, we propose a new strategy to suppress the influence of unimportant features (i.e., the features will be removed at the next pruning stage). Our comprehensive experiments on three benchmark datasets, i.e., CIFAR-10, ImageNet, and UCF-101, demonstrate the effectiveness of our CPLI method.

cs.CV↗

arXiv2019

Translating multispectral imagery to nighttime imagery via conditional generative adversarial networks

Nighttime satellite imagery has been applied in a wide range of fields. However, our limited understanding of how observed light intensity is formed and whether it can be simulated greatly hinders its further application. This study explores the potential of conditional Generative Adversarial Networks (cGAN) in translating multispectral imagery to nighttime imagery. A popular cGAN framework, pix2pix, was adopted and modified to facilitate this translation using gridded training image pairs derived from Landsat 8 and Visible Infrared Imaging Radiometer Suite (VIIRS). The results of this study prove the possibility of multispectral-to-nighttime translation and further indicate that, with the additional social media data, the generated nighttime imagery can be very similar to the ground-truth imagery. This study fills the gap in understanding the composition of satellite observed nighttime light and provides new paradigms to solve the emerging problems in nighttime remote sensing fields, including nighttime series construction, light desaturation, and multi-sensor calibration.

cs.CV↗

arXiv2019

IntersectGAN: Learning Domain Intersection for Generating Images with Multiple Attributes

Generative adversarial networks (GANs) have demonstrated great success in generating various visual content. However, images generated by existing GANs are often of attributes (e.g., smiling expression) learned from one image domain. As a result, generating images of multiple attributes requires many real samples possessing multiple attributes which are very resource expensive to be collected. In this paper, we propose a novel GAN, namely IntersectGAN, to learn multiple attributes from different image domains through an intersecting architecture. For example, given two image domains $X_1$ and $X_2$ with certain attributes, the intersection $X_1 \cap X_2$ denotes a new domain where images possess the attributes from both $X_1$ and $X_2$ domains. The proposed IntersectGAN consists of two discriminators $D_1$ and $D_2$ to distinguish between generated and real samples of different domains, and three generators where the intersection generator is trained against both discriminators. And an overall adversarial loss function is defined over three generators. As a result, our proposed IntersectGAN can be trained on multiple domains of which each presents one specific attribute, and eventually eliminates the need of real sample images simultaneously possessing multiple attributes. By using the CelebFaces Attributes dataset, our proposed IntersectGAN is able to produce high quality face images possessing multiple attributes (e.g., a face with black hair and a smiling expression). Both qualitative and quantitative evaluations are conducted to compare our proposed IntersectGAN with other baseline methods. Besides, several different applications of IntersectGAN have been explored with promising results.

cs.CV↗

arXiv2019

Refactoring Neural Networks for Verification

Deep neural networks (DNN) are growing in capability and applicability. Their effectiveness has led to their use in safety critical and autonomous systems, yet there is a dearth of cost-effective methods available for reasoning about the behavior of a DNN. In this paper, we seek to expand the applicability and scalability of existing DNN verification techniques through DNN refactoring. A DNN refactoring defines (a) the transformation of the DNN's architecture, i.e., the number and size of its layers, and (b) the distillation of the learned relationships between the input features and function outputs of the original to train the transformed network. Unlike with traditional code refactoring, DNN refactoring does not guarantee functional equivalence of the two networks, but rather it aims to preserve the accuracy of the original network while producing a simpler network that is amenable to more efficient property verification. We present an automated framework for DNN refactoring, and demonstrate its potential effectiveness through three case studies on networks used in autonomous systems.

cs.NE↗

arXiv2019

Hashing based Answer Selection

Answer selection is an important subtask of question answering (QA), where deep models usually achieve better performance. Most deep models adopt question-answer interaction mechanisms, such as attention, to get vector representations for answers. When these interaction based deep models are deployed for online prediction, the representations of all answers need to be recalculated for each question. This procedure is time-consuming for deep models with complex encoders like BERT which usually have better accuracy than simple encoders. One possible solution is to store the matrix representation (encoder output) of each answer in memory to avoid recalculation. But this will bring large memory cost. In this paper, we propose a novel method, called hashing based answer selection (HAS), to tackle this problem. HAS adopts a hashing strategy to learn a binary matrix representation for each answer, which can dramatically reduce the memory cost for storing the matrix representations of answers. Hence, HAS can adopt complex encoders like BERT in the model, but the online prediction of HAS is still fast with a low memory cost. Experimental results on three popular answer selection datasets show that HAS can outperform existing models to achieve state-of-the-art performance.

cs.CL↗

arXiv2019

Gated Group Self-Attention for Answer Selection

Answer selection (answer ranking) is one of the key steps in many kinds of question answering (QA) applications, where deep models have achieved state-of-the-art performance. Among these deep models, recurrent neural network (RNN) based models are most popular, typically with better performance than convolutional neural network (CNN) based models. Nevertheless, it is difficult for RNN based models to capture the information about long-range dependency among words in the sentences of questions and answers. In this paper, we propose a new deep model, called gated group self-attention (GGSA), for answer selection. GGSA is inspired by global self-attention which is originally proposed for machine translation and has not been explored in answer selection. GGSA tackles the problem of global self-attention that local and global information cannot be well distinguished. Furthermore, an interaction mechanism between questions and answers is also proposed to enhance GGSA by a residual structure. Experimental results on two popular QA datasets show that GGSA can outperform existing answer selection models to achieve state-of-the-art performance. Furthermore, GGSA can also achieve higher accuracy than global self-attention for the answer selection task, with a lower computation cost.

cs.CL↗

arXiv2019

Recent Advances in Transfer Learning for Cross-Dataset Visual Recognition: A Problem-Oriented Perspective

This paper takes a problem-oriented perspective and presents a comprehensive review of transfer learning methods, both shallow and deep, for cross-dataset visual recognition. Specifically, it categorises the cross-dataset recognition into seventeen problems based on a set of carefully chosen data and label attributes. Such a problem-oriented taxonomy has allowed us to examine how different transfer learning approaches tackle each problem and how well each problem has been researched to date. The comprehensive problem-oriented review of the advances in transfer learning with respect to the problem has not only revealed the challenges in transfer learning for visual recognition, but also the problems (e.g. eight of the seventeen problems) that have been scarcely studied. This survey not only presents an up-to-date technical review for researchers, but also a systematic approach and a reference for a machine learning practitioner to categorise a real problem and to look up for a possible solution accordingly.

cs.CV↗

arXiv2019

DVC: An End-to-end Deep Video Compression Framework

Conventional video compression approaches use the predictive coding architecture and encode the corresponding motion information and residual information. In this paper, taking advantage of both classical architecture in the conventional video compression method and the powerful non-linear representation ability of neural networks, we propose the first end-to-end video compression deep model that jointly optimizes all the components for video compression. Specifically, learning based optical flow estimation is utilized to obtain the motion information and reconstruct the current frames. Then we employ two auto-encoder style neural networks to compress the corresponding motion and residual information. All the modules are jointly learned through a single loss function, in which they collaborate with each other by considering the trade-off between reducing the number of compression bits and improving quality of the decoded video. Experimental results show that the proposed approach can outperform the widely used video coding standard H.264 in terms of PSNR and be even on par with the latest standard H.265 in terms of MS-SSIM. Code is released at https://github.com/GuoLusjtu/DVC.

eess.IV↗

arXiv2019

Observations of A Fast-Expanding and UV-Bright Type Ia Supernova SN 2013gs

In this paper, we present extensive optical and ultraviolet (UV) observations of the type Ia supernova (SN Ia) 2013gs discovered during the Tsinghua-NAOC Transient Survey. The photometric observations in the optical show that the light curves of SN 2013gs is similar to that of normal SNe Ia, with an absolute peak magnitude of $M_{B}$ = $-$19.25 $\pm$ 0.15 mag and a post-maximum decline rate $Δ$m$_{15}$(B) = 1.00 $ \pm $ 0.05 mag. \emph{Gehrels Swift} UVOT observations indicate that SN 2013gs shows unusually strong UV emission (especially in the $uvw1$ band) at around the maximum light (M$_{uvw1}$ $\sim$ $-$18.9 mag). The SN is characterized by relatively weak Fe~{\sc ii} {\sc iii} absorptions at $\sim$ 5000Å in the early spectra and a larger expansion velocity ($v_{Si}$ $\sim$ 13,000 km s$^{-1}$ around the maximum light) than the normal-velocity SNe Ia. We discuss the relation between the $uvw1-v$ color and some observables, including Si~{\sc ii} velocity, line strength of Si~{\sc ii} $λ$6355, Fe~{\sc ii}/{\sc iii} lines and $Δm_{15}$(B). Compared to other fast-expanding SNe Ia, SN 2013gs exhibits Si and Fe absorption lines with similar strength and bluer $uvw1-v$ color. We briefly discussed the origin of the observed UV dispersion of SNe Ia.

astro-ph.HE↗

arXiv2018

Learning Multi-level Deep Representations for Image Emotion Classification

In this paper, we propose a new deep network that learns multi-level deep representations for image emotion classification (MldrNet). Image emotion can be recognized through image semantics, image aesthetics and low-level visual features from both global and local views. Existing image emotion classification works using hand-crafted features or deep features mainly focus on either low-level visual features or semantic-level image representations without taking all factors into consideration. The proposed MldrNet combines deep representations of different levels, i.e. image semantics, image aesthetics, and low-level visual features to effectively classify the emotion types of different kinds of images, such as abstract paintings and web images. Extensive experiments on both Internet images and abstract paintings demonstrate the proposed method outperforms the state-of-the-art methods using deep features or hand-crafted features. The proposed approach also outperforms the state-of-the-art methods with at least 6% performance improvement in terms of overall classification accuracy.

cs.CV↗

arXiv2018

MUFold-BetaTurn: A Deep Dense Inception Network for Protein Beta-Turn Prediction

Beta-turn prediction is useful in protein function studies and experimental design. Although recent approaches using machine-learning techniques such as SVM, neural networks, and K-NN have achieved good results for beta-turn pre-diction, there is still significant room for improvement. As previous predictors utilized features in a sliding window of 4-20 residues to capture interactions among sequentially neighboring residues, such feature engineering may result in incomplete or biased features, and neglect interactions among long-range residues. Deep neural networks provide a new opportunity to address these issues. Here, we proposed a deep dense inception network (DeepDIN) for beta-turn prediction, which takes advantages of the state-of-the-art deep neural network design of the DenseNet and the inception network. A test on a recent BT6376 benchmark shows that the DeepDIN outperformed the previous best BetaTPred3 significantly in both the overall prediction accuracy and the nine-type beta-turn classification. A tool, called MUFold-BetaTurn, was developed, which is the first beta-turn prediction tool utilizing deep neural networks. The tool can be downloaded at http://dslsrv8.cs.missouri.edu/~cf797/MUFoldBetaTurn/download.html.

q-bio.BM↗

arXiv2018

Improving Protein Gamma-Turn Prediction Using Inception Capsule Networks

Protein gamma-turn prediction is useful in protein function studies and experimental design. Several methods for gamma-turn prediction have been developed, but the results were unsatisfactory with Matthew correlation coefficients (MCC) around 0.2-0.4. One reason for the low prediction accuracy is the limited capacity of the methods; in particular, the traditional machine-learning methods like SVM may not extract high-level features well to distinguish between turn or non-turn. Hence, it is worthwhile exploring new machine-learning methods for the prediction. A cutting-edge deep neural network, named Capsule Network (CapsuleNet), provides a new opportunity for gamma-turn prediction. Even when the number of input samples is relatively small, the capsules from CapsuleNet are very effective to extract high-level features for classification tasks. Here, we propose a deep inception capsule network for gamma-turn prediction. Its performance on the gamma-turn benchmark GT320 achieved an MCC of 0.45, which significantly outperformed the previous best method with an MCC of 0.38. This is the first gamma-turn prediction method utilizing deep neural networks. Also, to our knowledge, it is the first published bioinformatics application utilizing capsule network, which will provides a useful example for the community.

q-bio.QM↗

arXiv2018

The second closest gamma-ray burst: sub-luminous GRB 111005A with no supernova in a super-solar metallicity environment

We report the detection of the radio afterglow of a long gamma-ray burst (GRB) 111005A at 5-345 GHz, including the very long baseline interferometry observations with the positional error of 0.2 mas. The afterglow position is coincident with the disk of a galaxy ESO 580-49 at z= 0.01326 (~1" from its center), which makes GRB 111005A the second closest GRB known to date, after GRB 980425. The radio afterglow of GRB 111005A was an order of magnitude less luminous than those of local low-luminosity GRBs, and obviously than those of cosmological GRBs. The radio flux was approximately constant and then experienced an unusually rapid decay a month after the GRB explosion. Similarly to only two other GRBs, we did not find the associated supernovae (SN), despite deep near- and mid-infrared observations 1-9 days after the GRB explosion, reaching ~20 times fainter than other SNe associated with GRBs. Moreover, we measured twice solar metallicity for the GRB location. The low gamma-ray and radio luminosities, rapid decay, lack of a SN, and super-solar metallicity suggest that GRB 111005A represents a different rare class of GRBs than typical core-collapse events. We modelled the spectral energy distribution of the GRB 111005A host finding that it is a dwarf, moderately star-forming galaxy, similar to the host of GRB 980425. The existence of two local GRBs in such galaxies is still consistent with the hypothesis that the GRB rate is proportional to the cosmic star formation rate (SFR) density, but suggests that the GRB rate is biased towards low SFRs. Using the far-infrared detection of ESO 580-49, we conclude that the hosts of both GRBs 111005A and 980425 exhibit lower dust content than what would be expected from their stellar masses and optical colours.

astro-ph.HE↗

arXiv2017

A Monte Carlo approach to magnetar-powered transients: II. Broad-lined type Ic supernovae not associated with GRBs

Broad-lined type Ic supernovae (SNe Ic-BL) are a subclass of rare core collapse SNe whose energy source is debated in the literature. Recently a series of investigations on SNe Ic-BL with the magnetar (plus 56Ni) model were carried out. Evidence for magnetar formation was found for the well-observed SNe Ic-BL 1998bw and 2002ap. In this paper we systematically study a large sample of SNe Ic-BL not associated with gamma-ray bursts. We use photospheric velocity data determined in a homogeneous way. We find that the magnetar+56Ni model provides a good description of the light curves and velocity evolution of our sample of SNe Ic-BL, although some SNe (not all) can also be described by the pure-magnetar model or by the two-component pure-56Ni model (3 out of 12 are unlikely explained by two-component model). In the magnetar+56Ni model, the amount of 56Ni required to explain their luminosity is significantly reduced, and the derived initial explosion energy is, in general, in accordance with neutrino heating. Some correlations between different physical parameters are evaluated and their implications regarding magnetic field amplification and the total energy reservoir are discussed.

astro-ph.HE↗

arXiv2017

MUFold-SS: Protein Secondary Structure Prediction Using Deep Inception-Inside-Inception Networks

Motivation: Protein secondary structure prediction can provide important information for protein 3D structure prediction and protein functions. Deep learning, which has been successfully applied to various research fields such as image classification and voice recognition, provides a new opportunity to significantly improve the secondary structure prediction accuracy. Although several deep-learning methods have been developed for secondary structure prediction, there is room for improvement. MUFold-SS was developed to address these issues. Results: Here, a very deep neural network, the deep inception-inside-inception networks (Deep3I), is proposed for protein secondary structure prediction and a software tool was implemented using this network. This network takes two inputs: a protein sequence and a profile generated by PSI-BLAST. The output is the predicted eight states (Q8) or three states (Q3) of secondary structures. The proposed Deep3I not only achieves the state-of-the-art performance but also runs faster than other tools. Deep3I achieves Q3 82.8% and Q8 71.1% accuracies on the CB513 benchmark.

q-bio.QM↗