arXiv subjects

Peng Hu

Publications and source records attributed to Peng Hu.

At least 109 records · Page 6Linked to original sources

arXiv2023

Satellite Anomaly Detection Using Variance Based Genetic Ensemble of Neural Networks

In this paper, we use a variance-based genetic ensemble (VGE) of Neural Networks (NNs) to detect anomalies in the satellite's historical data. We use an efficient ensemble of the predictions from multiple Recurrent Neural Networks (RNNs) by leveraging each model's uncertainty level (variance). For prediction, each RNN is guided by a Genetic Algorithm (GA) which constructs the optimal structure for each RNN model. However, finding the model uncertainty level is challenging in many cases. Although the Bayesian NNs (BNNs)-based methods are popular for providing the confidence bound of the models, they cannot be employed in complex NN structures as they are computationally intractable. This paper uses the Monte Carlo (MC) dropout as an approximation version of BNNs. Then these uncertainty levels and each predictive model suggested by GA are used to generate a new model, which is then used for forecasting the TS and AD. Simulation results show that the forecasting and AD capability of the ensemble model outperforms existing approaches.

cs.LG↗

arXiv2023

Turbidity currents propagating down an inclined slope: particle auto-suspension

The Turbidity current (TC), a ubiquitous fluid-particle coupled phenomenon in the natural environment and engineering, can transport over long distances on an inclined terrain due to the suspension mechanism. A large-eddy simulation and discrete element method coupled model is employed to simulate the particle-laden gravity currents over the inclined slope in order to investigate the auto-suspension mechanism from a Lagrangian perspective. The particle Reynolds number in our TC simulation is $0.01\sim0.1$ and the slope angle is $1/20 \sim 1/5$. The influences of initial particle concentration and terrain slope on the particle flow regimes, particle movement patterns, fluid-particle interactions, energy budget and auto-suspension index are explored. The results indicate that the auto-suspension particles predominantly appear near the current head and their number increases and then decreases during the current evolution, which is positively correlated with the coherent structures around the head. When the turbidity current propagates downstream, the average particle Reynolds number of the auto-suspension particles remains basically unchanged, and is higher than that of other transported particles. The average particle Reynolds number of the transported particles exhibits a negative correlation with the Reynolds number of the current. Furthermore, the increase in particle concentration will enhance the particle velocity, which allows the turbidity current to advance faster and improves the perpendicular support, thereby increasing the turbidity current auto-suspension capacity. Increasing slope angle will result in a slightly larger front velocity, while the effect of that on the total force is insignificant.

physics.flu-dyn↗

arXiv2023

Incomplete Multi-view Clustering via Prototype-based Imputation

In this paper, we study how to achieve two characteristics highly-expected by incomplete multi-view clustering (IMvC). Namely, i) instance commonality refers to that within-cluster instances should share a common pattern, and ii) view versatility refers to that cross-view samples should own view-specific patterns. To this end, we design a novel dual-stream model which employs a dual attention layer and a dual contrastive learning loss to learn view-specific prototypes and model the sample-prototype relationship. When the view is missed, our model performs data recovery using the prototypes in the missing view and the sample-prototype relationship inherited from the observed view. Thanks to our dual-stream model, both cluster- and view-specific information could be captured, and thus the instance commonality and view versatility could be preserved to facilitate IMvC. Extensive experiments demonstrate the superiority of our method on six challenging benchmarks compared with 11 approaches. The code will be released.

cs.LG↗

arXiv2023

Laser Inter-Satellite Link Setup Delay: Quantification, Impact, and Tolerable Value

Dynamic laser inter-satellite links (LISLs) provide the flexibility of connecting a pair of satellites as required (dynamically) while static LISLs need to be active continuously between the energy-constrained satellites. However, due to the LISL establishment time (termed herein as LISL setup delay) being in the order of seconds, realizing dynamic LISLs is currently unfeasible. Towards the realization of dynamic LISLs, we first study the quantification of LISL setup delay; then we calculate the end-to-end latency of a free-space optical satellite network (FSOSN) with the LISL setup delay; subsequently, we analyze the impact of LISL setup delay on the end-to-end latency of the FSOSN. We also provide design guidelines for the laser communication terminal manufacturers in the form of maximum tolerable value of LISL setup delay for which the FSOSN based on Starlink's Phase I satellite constellation will be meaningful to use for low-latency long-distance inter-continental data communications.

cs.NI↗

arXiv2022

A Cross-Layer Descent Approach for Resilient Network Operations of Proliferated LEO Satellites

With the proliferated low-Earth-orbit (LEO) satellites in mega-constellations, the future Internet will be able to reach any place on Earth, providing high-quality services to everyone. However, high-quality operations in terms of timeliness and resilience are lacking in the current solutions. This paper proposes a multi-layer networking approach called "Cross-Layer Descent (CLD)". Based on the proposed system model, principles, and measures, CLD can support foundational services such as telecommand (TC) transmissions for various network operation missions for LEO satellites compliant with the Consultative Committee for Space Data Systems (CCSDS) standards. The CLD approach enhances timing and resilience requirements using advanced communication payloads. From the simulation-based analysis, the proposed scheme outperforms other classical ones in resilience and latency for typical TC missions. The future work and conclusive remarks are discussed at the end.

cs.NI↗

arXiv2022

Enabling Resilient and Real-Time Network Operations in Space: A Novel Multi-Layer Satellite Networking Scheme

Recently advanced low-Earth-orbit (LEO) satellite networks represented by large constellations and advanced payloads provide great promises for enabling high-quality Internet connectivity to any place on Earth. However, the traditional access-based approach to satellite operations cannot meet the pressing requirements of real-time, reliable, and resilient operations for LEO satellites. A new scheme is proposed based on multi-layer satellite networking considering the advanced Ka-band and optical communications payloads on a satellite platform. The proposed scheme can enable efficient and resilient message transmissions for critical telecommand and telemetry missions through different layers of satellite networks, which consist of LEO, medium-Earth-orbit (MEO), and geostationary (GEO) satellites. The proposed scheme is evaluated in a 24-hr satellite mission and shows superior performance improvements compared to the traditional operations approach.

eess.SY↗

arXiv2022

An Anomaly Detection Method for Satellites Using Monte Carlo Dropout

Recently, there has been a significant amount of interest in satellite telemetry anomaly detection (AD) using neural networks (NN). For AD purposes, the current approaches focus on either forecasting or reconstruction of the time series, and they cannot measure the level of reliability or the probability of correct detection. Although the Bayesian neural network (BNN)-based approaches are well known for time series uncertainty estimation, they are computationally intractable. In this paper, we present a tractable approximation for BNN based on the Monte Carlo (MC) dropout method for capturing the uncertainty in the satellite telemetry time series, without sacrificing accuracy. For time series forecasting, we employ an NN, which consists of several Long Short-Term Memory (LSTM) layers followed by various dense layers. We employ the MC dropout inside each LSTM layer and before the dense layers for uncertainty estimation. With the proposed uncertainty region and by utilizing a post-processing filter, we can effectively capture the anomaly points. Numerical results show that our proposed time series AD approach outperforms the existing methods from both prediction accuracy and AD perspectives.

cs.LG↗

arXiv2022

Multi-Scale Adaptive Network for Single Image Denoising

Multi-scale architectures have shown effectiveness in a variety of tasks thanks to appealing cross-scale complementarity. However, existing architectures treat different scale features equally without considering the scale-specific characteristics, \textit{i.e.}, the within-scale characteristics are ignored in the architecture design. In this paper, we reveal this missing piece for multi-scale architecture design and accordingly propose a novel Multi-Scale Adaptive Network (MSANet) for single image denoising. Specifically, MSANet simultaneously embraces the within-scale characteristics and the cross-scale complementarity thanks to three novel neural blocks, \textit{i.e.}, adaptive feature block (AFeB), adaptive multi-scale block (AMB), and adaptive fusion block (AFuB). In brief, AFeB is designed to adaptively preserve image details and filter noises, which is highly expected for the features with mixed details and noises. AMB could enlarge the receptive field and aggregate the multi-scale information, which meets the need of contextually informative features. AFuB devotes to adaptively sampling and transferring the features from one scale to another scale, which fuses the multi-scale features with varying characteristics from coarse to fine. Extensive experiments on both three real and six synthetic noisy image datasets show the superiority of MSANet compared with 12 methods. The code could be accessed from https://github.com/XLearning-SCU/2022-NeurIPS-MSANet.

eess.IV↗

arXiv2022

Bound States in the Continuum Based on the Total Internal Reflection of Bloch Waves

A photonic-crystal slab can support bound states in the continuum (BICs) which have infinite lifetimes but embedded into the continuous spectrum of optical modes in free space. The formation of BICs requires a total internal reflection (TIR) condition at both interfaces between the slab and free space. Here, we show that the TIR of Bloch waves can be directly obtained based on the generalized Fresnel equations proposed. If each of these Bloch waves picks up a phase with integer multiples of 2pi for traveling a round trip, light can be perfectly guided in the slab, namely, forming a BIC. A BIC solver with low computational complexity and fast convergence speed is developed, which can also work efficiently at high frequencies beyond the diffraction limit where multiple radiation channels exist. Two examples of multi-channel BICs are shown, and their topological nature in momentum space is also revealed. Both can be attributed to the coincidence of the topological charges of far-field radiations from different radiation channels. The concept of the generalized TIR and the TIR-based BIC solver developed offer highly effective approaches for explorations of BICs which could have many potential applications in guided-wave optics and enhanced light-matter interactions.

physics.optics↗

arXiv2022

Routing heterogeneous traffic in delay tolerant satellite networks

Delay Tolerant Networking (DTN) has been proposed as a new architecture to provide efficient store-carry-and-forward data transport in satellite networks. Since these networks relay on scheduled contact plans, the Contact Graph Routing (CGR) algorithm can be used to optimize routing and data delivery performance. However, in spite of the various improvements that have been made to CGR, there have been no significant proposals to prioritize traffic with different quality of service requirements. In this work we propose adaptations to CGR that allow performance improvements when sending traffic with different latency constraints, and develop a linear programming optimization model that works as a performance upper bound. The simulation results of the proposed schemes are promising and open the debate on other ways to improve performance while meeting the particular needs of heterogeneous traffic.

cs.NI↗

arXiv2022

KaraTuner: Towards end to end natural pitch correction for singing voice in karaoke

An automatic pitch correction system typically includes several stages, such as pitch extraction, deviation estimation, pitch shift processing, and cross-fade smoothing. However, designing these components with strategies often requires domain expertise and they are likely to fail on corner cases. In this paper, we present KaraTuner, an end-to-end neural architecture that predicts pitch curve and resynthesizes the singing voice directly from the tuned pitch and vocal spectrum extracted from the original recordings. Several vital technical points have been introduced in KaraTuner to ensure pitch accuracy, pitch naturalness, timbre consistency, and sound quality. A feed-forward Transformer is employed in the pitch predictor to capture longterm dependencies in the vocal spectrum and musical note. We also develop a pitch-controllable vocoder based on a novel source-filter block and the Fre-GAN architecture. KaraTuner obtains a higher preference than the rule-based pitch correction approach through A/B tests, and perceptual experiments show that the proposed vocoder achieves significant advantages in timbre consistency and sound quality compared with the parametric WORLD vocoder, phase vocoder and CLPC vocoder.

cs.SD↗

arXiv2022

Transcranial photoacoustic computed tomography of human brain function

Herein we report the first in-human transcranial imaging of brain function using photoacoustic computed tomography. Functional responses to benchmark motor tasks were imaged on both the skull-less and the skull-intact hemispheres of a hemicraniectomy patient. The observed brain responses in these preliminary results demonstrate the potential of photoacoustic computed tomography for achieving transcranial functional imaging.

physics.med-ph↗

arXiv2022

OPQ: Compressing Deep Neural Networks with One-shot Pruning-Quantization

As Deep Neural Networks (DNNs) usually are overparameterized and have millions of weight parameters, it is challenging to deploy these large DNN models on resource-constrained hardware platforms, e.g., smartphones. Numerous network compression methods such as pruning and quantization are proposed to reduce the model size significantly, of which the key is to find suitable compression allocation (e.g., pruning sparsity and quantization codebook) of each layer. Existing solutions obtain the compression allocation in an iterative/manual fashion while finetuning the compressed model, thus suffering from the efficiency issue. Different from the prior art, we propose a novel One-shot Pruning-Quantization (OPQ) in this paper, which analytically solves the compression allocation with pre-trained weight parameters only. During finetuning, the compression module is fixed and only weight parameters are updated. To our knowledge, OPQ is the first work that reveals pre-trained model is sufficient for solving pruning and quantization simultaneously, without any complex iterative/manual optimization at the finetuning stage. Furthermore, we propose a unified channel-wise quantization method that enforces all channels of each layer to share a common codebook, which leads to low bit-rate allocation without introducing extra overhead brought by traditional channel-wise quantization. Comprehensive experiments on ImageNet with AlexNet/MobileNet-V1/ResNet-50 show that our method improves accuracy and training efficiency while obtains significantly higher compression rates compared to the state-of-the-art.

cs.CV↗

arXiv2022

Fluid-particle interaction regimes during the evolution of turbidity currents from a coupled LES/DEM model

In this paper, fluid-particle interactions in lock-exchange turbidity currents (TCs) over a flat bed are investigated using a model combining LES and DEM (Large-Eddy Simulation and Discrete Element Method). The reliability of this model is demonstrated via comparing the numerical solutions with measurements of the front positions, fluid velocity profile, and particle concentration profile of lock-exchange TCs. The following physical understandings are obtained. The vorticity field plays an important role for the current evolution by affecting the fluid lift force (i.e., in the direction normal to the fluid-particle slip velocity) acting on the particles. At the very beginning, a longitudinal positive lift force due to strong positive vorticity promotes longitudinal particle transport. Afterwards, the longitudinal lift force decreases and eventually becomes negative, with a magnitude that even exceeds that of the positive longitudinal drag force, because more and more of the settling particles are affected by the negative vorticity near the bottom wall caused by surface friction. Interestingly, in spite of the complex behavior of the fluid-particle interaction forces and their role in TC evolution, only a very small fraction of the initial particle gravitational potential energy is actually transformed into TC kinetic energy (both particle and fluid).

physics.flu-dyn↗

arXiv2022

Closing the Management Gap for Satellite-Integrated Community Networks: A Hierarchical Approach to Self-Maintenance

Community networks (CNs) have become an important paradigm for providing essential Internet connectivity in unserved and underserved areas across the world. However, an indispensable part for CNs is network management, where responsive and autonomous maintenance is much needed. With the technological advancement in telecommunications networks, a classical satellite-dependent CN is envisioned to be transformed into a satellite-integrated CN (SICN), which will embrace significant autonomy, intelligence, and scalability in network management. This article discusses the machine-learning (ML) based hierarchical approach to enabling autonomous self-maintenance for SICNs. The approach is split into the anomaly identification and anomaly mitigation phases, where the related ML methods, data collection means, deployment options, and mitigation schemes are presented. With the case study, we discuss a typical scenario using satellite and fixed connections as backhaul options and show the effectiveness and performance improvements of the proposed approach with recurrent neural network and ensemble methods

cs.NI↗

arXiv2022

5G Enabled Fault Detection and Diagnostics: How Do We Achieve Efficiency?

The 5th-generation wireless networks (5G) technologies and mobile edge computing (MEC) provide great promises of enabling new capabilities for the industrial Internet of Things. However, the solutions enabled by the 5G ultra-reliable low-latency communication (URLLC) paradigm come with challenges, where URLLC alone does not necessarily guarantee the efficient execution of time-critical fault detection and diagnostics (FDD) applications. Based on the Tennessee Eastman Process model, we propose the concept of the communication-edge-computing (CEC) loop and a system model for evaluating the efficiency of FDD applications. We then formulate an optimization problem for achieving the defined CEC efficiency and discuss some typical solutions to the generic CEC-based FDD services, and propose a new uplink-based communication protocol called "ReFlexUp". From the performance analysis and numerical results, the proposed ReFlexUp protocol shows its effectiveness compared to the typical protocols such as Selective Repeat ARQ, HARQ, and "Occupy CoW" in terms of the key metrics such as latency, reliability, and efficiency. These results are further convinced from the mmWave-based simulations in a typical 5G MEC-based implementation.

cs.NI↗

arXiv2021

BRECQ: Pushing the Limit of Post-Training Quantization by Block Reconstruction

We study the challenging task of neural network quantization without end-to-end retraining, called Post-training Quantization (PTQ). PTQ usually requires a small subset of training data but produces less powerful quantized models than Quantization-Aware Training (QAT). In this work, we propose a novel PTQ framework, dubbed BRECQ, which pushes the limits of bitwidth in PTQ down to INT2 for the first time. BRECQ leverages the basic building blocks in neural networks and reconstructs them one-by-one. In a comprehensive theoretical study of the second-order error, we show that BRECQ achieves a good balance between cross-layer dependency and generalization error. To further employ the power of quantization, the mixed precision technique is incorporated in our framework by approximating the inter-layer and intra-layer sensitivity. Extensive experiments on various handcrafted and searched neural architectures are conducted for both image classification and object detection tasks. And for the first time we prove that, without bells and whistles, PTQ can attain 4-bit ResNet and MobileNetV2 comparable with QAT and enjoy 240 times faster production of quantized models. Codes are available at https://github.com/yhhhli/BRECQ.

cs.LG↗

arXiv2021

Unsupervised Neural Rendering for Image Hazing

Image hazing aims to render a hazy image from a given clean one, which could be applied to a variety of practical applications such as gaming, filming, photographic filtering, and image dehazing. To generate plausible haze, we study two less-touched but challenging problems in hazy image rendering, namely, i) how to estimate the transmission map from a single image without auxiliary information, and ii) how to adaptively learn the airlight from exemplars, i.e., unpaired real hazy images. To this end, we propose a neural rendering method for image hazing, dubbed as HazeGEN. To be specific, HazeGEN is a knowledge-driven neural network which estimates the transmission map by leveraging a new prior, i.e., there exists the structure similarity (e.g., contour and luminance) between the transmission map and the input clean image. To adaptively learn the airlight, we build a neural module based on another new prior, i.e., the rendered hazy image and the exemplar are similar in the airlight distribution. To the best of our knowledge, this could be the first attempt to deeply rendering hazy images in an unsupervised fashion. Comparing with existing haze generation methods, HazeGEN renders the hazy images in an unsupervised, learnable, and controllable manner, thus avoiding the labor-intensive efforts in paired data collection and the domain-shift issue in haze generation. Extensive experiments show the promising performance of our method comparing with some baselines in both qualitative and quantitative comparisons. The code will be released on GitHub after acceptance.

cs.CV↗