arXiv subjects

Le Yang

Publications and source records attributed to Le Yang.

At least 73 records · Page 4Linked to original sources

arXiv2022

Analysis and Optimization of Hybrid Caching in mmWave Networks with BS Cooperation

In this paper, we investigate a hybrid caching strategy maximizing the success transmission probability (STP) in a millimeter wave (mmWave) cache-enabled network. First, we derive theoretical expressions of the STP and the average system transmission delay by utilizing stochastic geometry, then we consider the maximization of the STP and the minimization of the average system transmission delay by optimizing the design parameters. Considering the optimality structure of the NP-hard problem, the original problem is transferred into a multi-choice knapsack problem (MCKP). Finally, we investigate the impact of key network parameters on the STP and the average system transmission delay. Numerical results demonstrate the superiority of the proposed caching strategy over the conventional caching strategies in the mmWave cache-enabled networks.

cs.IT↗

arXiv2022

Spatio-Temporal Analysis of SINR Meta Distribution for mmWave Heterogeneous Networks Under Geo/G/1 Queues

A fine-grained analysis of network performance is crucial for system design. In this paper, we focus on the meta distribution of the signal-to-interference-plus-noise-ratio (SINR) in the mmWave heterogeneous networks where the base stations (BS) in each tier are modeled as a Poisson point process (PPP). By utilizing stochastic geometry and queueing theory, we characterize the spatial and temporal randomness while the special characteristics of mmWave communications, including different path loss laws for line-of-sight and non-line-of-sight links and directional beamforming, are incorporated into the analysis. We derive the moments of the conditional successful transmission probability (STP). By taking the temporal random arrival of traffic into consideration, an equation is formulated to derive the meta distribution and the meta distribution can be obtained in a recursive manner. The numerical results reveal the impact of the key network parameters, such as the SINR threshold and the blockage parameter, on the network performance.

cs.IT↗

arXiv2021

Dynamic Neural Networks: A Survey

Dynamic neural network is an emerging research topic in deep learning. Compared to static models which have fixed computational graphs and parameters at the inference stage, dynamic networks can adapt their structures or parameters to different inputs, leading to notable advantages in terms of accuracy, computational efficiency, adaptiveness, etc. In this survey, we comprehensively review this rapidly developing area by dividing dynamic networks into three main categories: 1) instance-wise dynamic models that process each instance with data-dependent architectures or parameters; 2) spatial-wise dynamic networks that conduct adaptive computation with respect to different spatial locations of image data and 3) temporal-wise dynamic models that perform adaptive inference along the temporal dimension for sequential data such as videos and texts. The important research problems of dynamic networks, e.g., architecture design, decision making scheme, optimization technique and applications, are reviewed systematically. Finally, we discuss the open problems in this field together with interesting future research directions.

cs.CV↗

arXiv2021

Background-Click Supervision for Temporal Action Localization

Weakly supervised temporal action localization aims at learning the instance-level action pattern from the video-level labels, where a significant challenge is action-context confusion. To overcome this challenge, one recent work builds an action-click supervision framework. It requires similar annotation costs but can steadily improve the localization performance when compared to the conventional weakly supervised methods. In this paper, by revealing that the performance bottleneck of the existing approaches mainly comes from the background errors, we find that a stronger action localizer can be trained with labels on the background video frames rather than those on the action frames. To this end, we convert the action-click supervision to the background-click supervision and develop a novel method, called BackTAL. Specifically, BackTAL implements two-fold modeling on the background video frames, i.e. the position modeling and the feature modeling. In position modeling, we not only conduct supervised learning on the annotated video frames but also design a score separation module to enlarge the score differences between the potential action frames and backgrounds. In feature modeling, we propose an affinity module to measure frame-specific similarities among neighboring frames and dynamically attend to informative neighbors when calculating temporal convolution. Extensive experiments on three benchmarks are conducted, which demonstrate the high performance of the established BackTAL and the rationality of the proposed background-click supervision. Code is available at https://github.com/VividLe/BackTAL.

cs.CV↗

arXiv2021

Enhanced Medium Range Order in Vapor Deposited Germania Glasses at Elevated Temperatures

Glasses are nonequilibrium solids with properties highly dependent on their method of preparation. In vapor-deposited molecular glasses, structural organization could be readily tuned with deposition rate and substrate temperature. Herein, we show the atomic arrangement of strong network forming GeO2 glass is modified at medium range (< 2 nm) through vapor deposition at elevated temperatures. Raman spectral signatures distinctively show the population of 6-membered GeO4 rings increases at elevated substrate temperatures. Deposition near the glass transition temperature is more efficient than post-growth annealing in modifying atomic structure at medium range. The enhanced medium range organization correlates with reduction of the room temperature internal friction. Identifying the microscopic origin of room temperature internal friction in amorphous oxides is paramount to design the next generation interference coatings for mirrors of the end test masses of gravitational wave interferometers, in which the room temperature internal friction is a main source of noise limiting their sensitivity.

cond-mat.mtrl-sci↗

arXiv2021

Low mechanical loss TiO$_2$:GeO$_2$ coatings for reduced thermal noise in Gravitational Wave Interferometers

The sensitivity of current and planned gravitational wave interferometric detectors is limited, in the most critical frequency region around 100 Hz, by a combination of quantum noise and thermal noise. The latter is dominated by Brownian noise: thermal motion originating from the elastic energy dissipation in the dielectric coatings used in the interferometer mirrors. The energy dissipation is a material property characterized by the mechanical loss angle. We have identified mixtures of titanium dioxide (TiO$_2$) and germanium dioxide (GeO$_2$) that show internal dissipations at a level of 1 $\times 10^{-4}$, low enough to provide almost a factor of two improvement on the level of Brownian noise with respect to the state-of-the-art materials. We show that by using a mixture of 44% TiO$_2$ and 56% GeO$_2$ in the high refractive index layers of the interferometer mirrors, it would be possible to achieve a thermal noise level in line with the design requirements. These results are a crucial step forward to produce the mirrors needed to meet the thermal noise requirements for the planned upgrades of the Advanced LIGO and Virgo detectors.

physics.ins-det↗

arXiv2021

Comprehensive study of amorphous metal oxide and Ta$_2$O$_5$-based mixed oxide coatings for gravitational-wave detectors

High finesse optical cavities of current interferometric gravitational-wave detectors are significantly limited in sensitivity by laser quantum noise and coating thermal noise. The thermal noise is associated with internal energy dissipation in the materials that compose the test masses of the interferometer. Our understanding of how the internal friction is linked to the amorphous material structure is limited due to the complexity of the problem and the lack of studies that span over a large range of materials. We present a systematic investigation of amorphous metal oxide and Ta$_2$O$_5$-based mixed oxide coatings to evaluate their suitability for low Brownian noise experiments. It is shown that the mechanical loss of metal oxides is correlated to their amorphous morphology, with continuous random network materials such as SiO$_2$ and GeO$_2$ featuring the lowest loss angles. We evaluated different Ta$_2$O$_5$-based mixed oxide thin films and studied the influence of the dopant in the optical and elastic properties of the coating. We estimated the thermal noise associated with high-reflectance multilayer stacks that employ each of the mixed oxides as the high index material. We concluded that the current high index material of TiO$_2$-doped Ta$_2$O$_5$ is the optimal choice for reduced thermal noise among Ta$_2$O$_5$-based mixed oxide coatings with low dopant concentrations.

cond-mat.mtrl-sci↗

arXiv2021

Coverage Probability and Energy Efficiency of Reconfigurable Intelligent Surface-Assisted mmWave Networks

Reconfigurable intelligent surface (RIS) emerges as a promising technology for the next generation networks. In this paper, we utilize the tools from stochastic geometry to study the performance of a RIS-assisted millimeter wave (mmWave) cellular network. Specifically, the locations of the base stations (BSs) and the midpoints of the blockages are modeled as two independent Poisson point processes (PPP) where the blockages are modeled by line boolean model and a fraction of the blockages are coated with RISs. The distinguish characteristics of mmWave communications, i.e., the directional beamforming and different path loss laws for line-of-sight (LOS) and non-line-of-sight (NLOS), are incorporated into the analysis. We derive the expressions of the coverage probability and the area spectral efficiency. The coverage probability under the special case where the blockage parameter is sufficiently small is also derived. Numerical results demonstrate that better coverage performance and higher energy efficiency can be achieved by the large-scale deployment of RISs. In addition, the tradeoff between the BS and RIS densities is investigated and the results show that the RISs are excellent supplementary for the traditional networks to improve the coverage probability with limited power consumption.

cs.IT↗

arXiv2021

CondenseNet V2: Sparse Feature Reactivation for Deep Networks

Reusing features in deep networks through dense connectivity is an effective way to achieve high computational efficiency. The recent proposed CondenseNet has shown that this mechanism can be further improved if redundant features are removed. In this paper, we propose an alternative approach named sparse feature reactivation (SFR), aiming at actively increasing the utility of features for reusing. In the proposed network, named CondenseNetV2, each layer can simultaneously learn to 1) selectively reuse a set of most important features from preceding layers; and 2) actively update a set of preceding features to increase their utility for later layers. Our experiments show that the proposed models achieve promising performance on image classification (ImageNet and CIFAR) and object detection (MS COCO) in terms of both theoretical efficiency and practical speed.

cs.CV↗

arXiv2021

On Meta Distribution and Local Delay for Cache-Enabled Networks with Random DTX: Analysis and Optimization

A fine-grained analysis of the cache-enabled networks is crucial for system design. In this paper, we focus on the meta distribution of the signal-to-interference ratio (SIR) for the cache-enabled networks where the locations of the base stations (BSs) are modeled as a Poisson point process (PPP). With the application of the random caching and the random discontinuous transmission (DTX) schemes, we derive the moments of the conditional successful transmission probability (STP), the exact meta distribution and its beta approximation by utilizing stochastic geometry. The closed-form expressions of the mean and variance of the local delay (i.e., the network jitter) are also derived. We then consider the maximization of the mean STP and the minimization of the average system transmission delay by jointly optimizing the caching probability and the BS active probability. Finally, the numerical results demonstrate the superiority of the proposed optimization schemes over the existing caching strategies and reveal the impacts of the key network parameters on the cache-enabled networks in terms of mean STP, STP variance, meta distribution, mean local delay and network jitter.

cs.IT↗

arXiv2021

Revisiting Locally Supervised Learning: an Alternative to End-to-end Training

Due to the need to store the intermediate activations for back-propagation, end-to-end (E2E) training of deep networks usually suffers from high GPUs memory footprint. This paper aims to address this problem by revisiting the locally supervised learning, where a network is split into gradient-isolated modules and trained with local supervision. We experimentally show that simply training local modules with E2E loss tends to collapse task-relevant information at early layers, and hence hurts the performance of the full model. To avoid this issue, we propose an information propagation (InfoPro) loss, which encourages local modules to preserve as much useful information as possible, while progressively discard task-irrelevant information. As InfoPro loss is difficult to compute in its original form, we derive a feasible upper bound as a surrogate optimization objective, yielding a simple but effective algorithm. In fact, we show that the proposed method boils down to minimizing the combination of a reconstruction loss and a normal cross-entropy/contrastive term. Extensive empirical results on five datasets (i.e., CIFAR, SVHN, STL-10, ImageNet and Cityscapes) validate that InfoPro is capable of achieving competitive performance with less than 40% memory footprint compared to E2E training, while allowing using training data with higher-resolution or larger batch sizes under the same GPU memory constraint. Our method also enables training local modules asynchronously for potential training acceleration. Code is available at: https://github.com/blackfeather-wang/InfoPro-Pytorch.

cs.CV↗

arXiv2020

Glance and Focus: a Dynamic Approach to Reducing Spatial Redundancy in Image Classification

The accuracy of deep convolutional neural networks (CNNs) generally improves when fueled with high resolution images. However, this often comes at a high computational cost and high memory footprint. Inspired by the fact that not all regions in an image are task-relevant, we propose a novel framework that performs efficient image classification by processing a sequence of relatively small inputs, which are strategically selected from the original image with reinforcement learning. Such a dynamic decision process naturally facilitates adaptive inference at test time, i.e., it can be terminated once the model is sufficiently confident about its prediction and thus avoids further redundant computation. Notably, our framework is general and flexible as it is compatible with most of the state-of-the-art light-weighted CNNs (such as MobileNets, EfficientNets and RegNets), which can be conveniently deployed as the backbone feature extractor. Experiments on ImageNet show that our method consistently improves the computational efficiency of a wide variety of deep models. For example, it further reduces the average latency of the highly efficient MobileNet-V3 on an iPhone XS Max by 20% without sacrificing accuracy. Code and pre-trained models are available at https://github.com/blackfeather-wang/GFNet-Pytorch.

cs.CV↗

arXiv2020

Structural evolution of binary oxide nanolaminates with annealing and its impact on room-temperature internal friction

Internal friction in oxide thin films imposes a critical limitation to the sensitivity and stability of ultra-high finesse optical cavities for gravitational wave detectors. Strategies like doping or creating nanolaminates are sought to introduce structural modifications that reduce internal friction. This work describes an investigation of the morphological changes SiO2/Ta2O5 and TiO2/Ta2O5 nanolaminates undergo with annealing and their impact on room temperature internal friction. It is demonstrated that thermal treatment results in a reduction of internal friction in both nanolaminates, but through different pathways. In the SiO2/Ta2O5 nanolaminate, which layers remain intact after annealing, the total reduction in internal friction follows the reduction in the composing SiO2 and Ta2O5 layers. Instead, interdiffusion initiated by annealing at the interface of the TiO2/Ta2O5 nanolaminate and the formation of a mixed phase dictate a more significant reduction in internal friction to ~ 2.6 * 10-4, a value lower than any other Ta2O5 mixture coating with similar cation concentration.

cond-mat.mtrl-sci↗

arXiv2020

Prediction of crystallized phases of amorphous Ta$_2$O$_5$-based mixed oxide thin films using density functional theory calculations

The genomics approach to materials, heralded by increasingly accurate density functional theory (DFT) calculations conducted on thousands of crystalline compounds, has led to accelerated material discovery and property predictions. However, so far amorphous materials have been largely excluded from this as these systems are notoriously difficult to simulate. Here we study amorphous Ta$_2$O$_5$ thin films mixed with Al$_2$O$_3$, SiO$_2$, Sc$_2$O$_3$, TiO$_2$, ZnO, ZrO$_2$, Nb$_2$O$_5$ and HfO$_2$ to identify their crystalline structure upon post-deposition annealing in air both experimentally and with simulations. Using the Materials Project open database, phase diagrams based on DFT calculations are constructed for the mixed oxide systems and the annealing process is evaluated via grand potential diagrams with varying oxygen chemical potential. Despite employing calculations based on crystalline bulk materials, the predictions agree well with the experimentally observed crystallized phases of the amorphous films. Only in two cases the database leads to incorrect predictions: in TiO$_2$-doped Ta$_2$O$_5$ because it does not contain a ternary compound found experimentally, and in Sc$_2$O$_3$-doped Ta$_2$O$_5$ because DFT overestimates the formation enthalpy difference between Sc$_2$O$_3$ and Ta$_2$O$_5$ and thus does not reproduce observed oxygen competition effects. In the absence of ternary phases, the dopant acts as an amorphizer agent increasing the thermal stability of Ta$_2$O$_5$. These results show that DFT calculations can be applied for the prediction of crystallized structures of annealed amorphous materials. This could pave the way for accelerated \textit{in silico} material discovery and property predictions using the powerful genomic approach for amorphous oxide coatings employed in a wide range of applications such as optical coatings, energy storage and electronic devices.

cond-mat.mtrl-sci↗

arXiv2020

Equivalent Classification Mapping for Weakly Supervised Temporal Action Localization

Weakly supervised temporal action localization is a newly emerging yet widely studied topic in recent years. The existing methods can be categorized into two localization-by-classification pipelines, i.e., the pre-classification pipeline and the post-classification pipeline. The pre-classification pipeline first performs classification on each video snippet and then aggregate the snippet-level classification scores to obtain the video-level classification score. In contrast, the post-classification pipeline aggregates the snippet-level features first and then predicts the video-level classification score based on the aggregated feature. Although the classifiers in these two pipelines are used in different ways, the role they play is exactly the same---to classify the given features to identify the corresponding action categories. To this end, an ideal classifier can make both pipelines work. This inspires us to simultaneously learn these two pipelines in a unified framework to obtain an effective classifier. Specifically, in the proposed learning framework, we implement two parallel network streams to model the two localization-by-classification pipelines simultaneously and make the two network streams share the same classifier. This achieves the novel Equivalent Classification Mapping (ECM) mechanism. Moreover, we discover that an ideal classifier may possess two characteristics: 1) The frame-level classification scores obtained from the pre-classification stream and the feature aggregation weights in the post-classification stream should be consistent; 2) The classification results of these two streams should be identical. Based on these two characteristics, we further introduce a weight-transition module and an equivalent training strategy into the proposed learning framework, which assists to thoroughly mine the equivalence mechanism.

cs.CV↗

arXiv2020

A Fine-Grained Analysis of mmWave Heterogeneous Networks

A fine-grained analysis of the cache-enabled networks is crucial for system design. In this paper, we focus on the meta distribution of the signal-to-interference-plus-noise-ratio (SINR) in the mmWave heterogeneous networks where the base stations (BS) in each tier are modeled as Poisson point process (PPP). By utilizing stochastic geometry, we derive the moments of the conditional success probability, based on which the exact expression of meta distribution and its beta approximation are derived. In addition, key performance metrics, the success probability, the variance of the conditional success probability, the mean local delay and the network jitter are achieved. The distinguishing characteristics of the mmWave communications, including different path loss laws for line-of-sight and non-line-of-sight links and directional beamforming are incorporated into the analysis. The simulation results reveal the impact of the key network parameters, such as blockage parameter, bias factor, number of antenna elements and density on the performance.

cs.IT↗

arXiv2020

Revisiting Anchor Mechanisms for Temporal Action Localization

Most of the current action localization methods follow an anchor-based pipeline: depicting action instances by pre-defined anchors, learning to select the anchors closest to the ground truth, and predicting the confidence of anchors with refinements. Pre-defined anchors set prior about the location and duration for action instances, which facilitates the localization for common action instances but limits the flexibility for tackling action instances with drastic varieties, especially for extremely short or extremely long ones. To address this problem, this paper proposes a novel anchor-free action localization module that assists action localization by temporal points. Specifically, this module represents an action instance as a point with its distances to the starting boundary and ending boundary, alleviating the pre-defined anchor restrictions in terms of action localization and duration. The proposed anchor-free module is capable of predicting the action instances whose duration is either extremely short or extremely long. By combining the proposed anchor-free module with a conventional anchor-based module, we propose a novel action localization framework, called A2Net. The cooperation between anchor-free and anchor-based modules achieves superior performance to the state-of-the-art on THUMOS14 (45.5% vs. 42.8%). Furthermore, comprehensive experiments demonstrate the complementarity between the anchor-free and the anchor-based module, making A2Net simple but effective.

cs.CV↗

arXiv2020

Spatio-Temporal Analysis of Cellular Networks with Cell-Center/Edge Users

Emergence of various types of services has brought about explosive growth of traffic as well as diversified traffic characteristics in the cellular networks. To have a comprehensive understanding of the influences caused by various traffic status is vital for the deployment of the next-generation wireless networks. In this paper, we develop a mathematical analytical model by utilizing queuing theory and stochastic geometry where the randomness of the traffic and the geographical locations of the interferers can be captured. We derive the b-th moments of the conditional success probability and the closed-form expressions of the meta distribution for the cell-center users (CCUs) and the cell-edge users (CEUs), respectively. Fixed-point equations are then formulated to obtain the exact value of the meta distribution by taking the random arrival traffic into consideration and the impact of the random arrival traffic on the queue status is revealed. In addition, the mean local delays for CCUs nad CEUs are derived and the corresponding regions for CCUs and CEUs where the mean local delays maintain finite are obtained. Finally, the impact of the critical network parameters on the meta distribution and the mean local delay is investigated with the numerical results.

cs.IT↗