arXiv subjects

Kai Huang

Publications and source records attributed to Kai Huang.

At least 163 records · Page 9Linked to original sources

arXiv2021

Visible and Near-infrared Microdisk Resonators on a 4H-Silicon-Carbide-on-Insulator Platform

Wavelength-sized microdisk resonators were fabricated on a single crystalline 4H-silicon-carbide-oninsulator platform (4H-SiCOI). By carrying out microphotoluminescence measurements at room temperature, we show that the microdisk resonators support whispering-gallery modes (WGMs) with quality factors up to $5.25 \times 10^3$ and mode volumes down to $2.69 \times(λ/n)^3$ at the visible and near-infrared wavelengths. Moreover, the demonstrated wavelength-sized microdisk resonators exhibit WGMs whose resonant wavelengths compatible with the zero-phonon lines of spin defects in 4H-SiCOI, making them a promising candidate for applications in cavity quantum electrodynamics and integrated quantum photonic circuits.

physics.optics↗

arXiv2020

Context-Aware Drive-thru Recommendation Service at Fast Food Restaurants

Drive-thru is a popular sales channel in the fast food industry where consumers can make food purchases without leaving their cars. Drive-thru recommendation systems allow restaurants to display food recommendations on the digital menu board as guests are making their orders. Popular recommendation models in eCommerce scenarios rely on user attributes (such as user profiles or purchase history) to generate recommendations, while such information is hard to obtain in the drive-thru use case. Thus, in this paper, we propose a new recommendation model Transformer Cross Transformer (TxT), which exploits the guest order behavior and contextual features (such as location, time, and weather) using Transformer encoders for drive-thru recommendations. Empirical results show that our TxT model achieves superior results in Burger King's drive-thru production environment compared with existing recommendation solutions. In addition, we implement a unified system to run end-to-end big data analytics and deep learning workloads on the same cluster. We find that in practice, maintaining a single big data cluster for the entire pipeline is more efficient and cost-saving. Our recommendation system is not only beneficial for drive-thru scenarios, and it can also be generalized to other customer interaction channels.

cs.IR↗

arXiv2020

3D Object Detection and Tracking Based on Streaming Data

Recent approaches for 3D object detection have made tremendous progresses due to the development of deep learning. However, previous researches are mostly based on individual frames, leading to limited exploitation of information between frames. In this paper, we attempt to leverage the temporal information in streaming data and explore 3D streaming based object detection as well as tracking. Toward this goal, we set up a dual-way network for 3D object detection based on keyframes, and then propagate predictions to non-key frames through a motion based interpolation algorithm guided by temporal information. Our framework is not only shown to have significant improvements on object detection compared with frame-by-frame paradigm, but also proven to produce competitive results on KITTI Object Tracking Benchmark, with 76.68% in MOTA and 81.65% in MOTP respectively.

cs.CV↗

arXiv2020

Complex Robotic Manipulation via Graph-Based Hindsight Goal Generation

Reinforcement learning algorithms such as hindsight experience replay (HER) and hindsight goal generation (HGG) have been able to solve challenging robotic manipulation tasks in multi-goal settings with sparse rewards. HER achieves its training success through hindsight replays of past experience with heuristic goals, but under-performs in challenging tasks in which goals are difficult to explore. HGG enhances HER by selecting intermediate goals that are easy to achieve in the short term and promising to lead to target goals in the long term. This guided exploration makes HGG applicable to tasks in which target goals are far away from the object's initial position. However, HGG is not applicable to manipulation tasks with obstacles because the euclidean metric used for HGG is not an accurate distance metric in such environments. In this paper, we propose graph-based hindsight goal generation (G-HGG), an extension of HGG selecting hindsight goals based on shortest distances in an obstacle-avoiding graph, which is a discrete representation of the environment. We evaluated G-HGG on four challenging manipulation tasks with obstacles, where significant enhancements in both sample efficiency and overall success rate are shown over HGG and HER. Videos can be viewed at https://sites.google.com/view/demos-g-hgg/.

cs.RO↗

arXiv2020

Large-scale Uncertainty Estimation and Its Application in Revenue Forecast of SMEs

The economic and banking importance of the small and medium enterprise (SME) sector is well recognized in contemporary society. Business credit loans are very important for the operation of SMEs, and the revenue is a key indicator of credit limit management. Therefore, it is very beneficial to construct a reliable revenue forecasting model. If the uncertainty of an enterprise's revenue forecasting can be estimated, a more proper credit limit can be granted. Natural gradient boosting approach, which estimates the uncertainty of prediction by a multi-parameter boosting algorithm based on the natural gradient. However, its original implementation is not easy to scale into big data scenarios, and computationally expensive compared to state-of-the-art tree-based models (such as XGBoost). In this paper, we propose a Scalable Natural Gradient Boosting Machines that is simple to implement, readily parallelizable, interpretable and yields high-quality predictive uncertainty estimates. According to the characteristics of revenue distribution, we derive an uncertainty quantification function. We demonstrate that our method can distinguish between samples that are accurate and inaccurate on revenue forecasting of SMEs. What's more, interpretability can be naturally obtained from the model, satisfying the financial needs.

cs.LG↗

arXiv2020

Radar for projectile impact on granular media

From the prevention of natural disasters such as landslide and avalanches, to the enhancement of energy efficiencies in chemical and civil engineering industries, understanding the collective dynamics of granular materials is a fundamental question that are closely related to our daily lives. Using a recently developed multi-static radar system operating at $10$\,GHz (X-band), we explore the possibility of tracking a projectile moving inside a granular medium, focusing on possible sources of uncertainties in the detection and reconstruction processes. On the one hand, particle tracking with continuous wave radar provides an extremely high temporal resolution. On the other hand, there are still challenges in obtaining tracer trajectories accurately. We show that some of the challenges can be resolved through a correction of the IQ mismatch in the raw signals obtained. Consequently, the tracer trajectories can be obtained with sub-millimeter spatial resolution. Such an advance can not only shed light on radar particle tracking, but also on a wide range of scenarios where issues relevant to IQ mismatch arise.

physics.app-ph↗

arXiv2020

Indirect and Direct Training of Spiking Neural Networks for End-to-End Control of a Lane-Keeping Vehicle

Building spiking neural networks (SNNs) based on biological synaptic plasticities holds a promising potential for accomplishing fast and energy-efficient computing, which is beneficial to mobile robotic applications. However, the implementations of SNNs in robotic fields are limited due to the lack of practical training methods. In this paper, we therefore introduce both indirect and direct end-to-end training methods of SNNs for a lane-keeping vehicle. First, we adopt a policy learned using the \textcolor{black}{Deep Q-Learning} (DQN) algorithm and then subsequently transfer it to an SNN using supervised learning. Second, we adopt the reward-modulated spike-timing-dependent plasticity (R-STDP) for training SNNs directly, since it combines the advantages of both reinforcement learning and the well-known spike-timing-dependent plasticity (STDP). We examine the proposed approaches in three scenarios in which a robot is controlled to keep within lane markings by using an event-based neuromorphic vision sensor. We further demonstrate the advantages of the R-STDP approach in terms of the lateral localization accuracy and training time steps by comparing them with other three algorithms presented in this paper.

cs.NE↗

arXiv2020

Set Voronoi Tessellation for Particulate Systems in Two Dimensions

Given a countable set of points in a continuous space, Voronoi tessellation is an intuitive way of partitioning the space according to the distance to the individual points. As a powerful approach to obtain structural information, it has a long history and widespread applications in diverse disciplines, from astronomy to urban planning. For particulate systems in real life, such as a pile of sand or a crowd of pedestrians, the realization of Voronoi tessellation needs to be modified to accommodate the fact that the particles cannot be simply treated as points. Here, we elucidate the use of Set Voronoi tessellation (i.e., considering for a non-spherical particle a set of points on its surface) to extract meaningful local information in a quasi-two-dimensional system of granular rods. In addition, we illustrate how it can be applied to arbitrarily shaped particles such as an assembly of honey bees or pedestrians for obtaining structural information. Details on the implementation of this algorithm with the strategy of balancing computational cost and accuracy are discussed. Furthermore, we provide our python code as open source in order to facilitate Set Voronoi calculations in two dimensions for arbitrarily shaped objects.

cond-mat.soft↗

arXiv2019

Scaling of magnetic reconnection with a limited x-line extent

Contrary to all the 2D models, where the reconnection x-line extent is infinitely long, we study magnetic reconnection in the opposite limit. The scaling of the average reconnection rate and outflow speed are modeled as a function of the x-line extent. An internal x-line asymmetry along the current direction develops because of the flux transport by electrons beneath the ion kinetic scale, and it plays an important role in suppressing reconnection in the short x-line limit; the average reconnection rate drops because of the limited active region, and the outflow speed reduction is associated with the reduction of the $J \times B$ force, that is caused by the phase shift between the J and B profiles, also as a consequence of this flux transport.

physics.space-ph↗

arXiv2019

PhoneBit: Efficient GPU-Accelerated Binary Neural Network Inference Engine for Mobile Phones

Over the last years, a great success of deep neural networks (DNNs) has been witnessed in computer vision and other fields. However, performance and power constraints make it still challenging to deploy DNNs on mobile devices due to their high computational complexity. Binary neural networks (BNNs) have been demonstrated as a promising solution to achieve this goal by using bit-wise operations to replace most arithmetic operations. Currently, existing GPU-accelerated implementations of BNNs are only tailored for desktop platforms. Due to architecture differences, mere porting of such implementations to mobile devices yields suboptimal performance or is impossible in some cases. In this paper, we propose PhoneBit, a GPU-accelerated BNN inference engine for Android-based mobile devices that fully exploits the computing power of BNNs on mobile GPUs. PhoneBit provides a set of operator-level optimizations including locality-friendly data layout, bit packing with vectorization and layers integration for efficient binary convolution. We also provide a detailed implementation and parallelization optimization for PhoneBit to optimally utilize the memory bandwidth and computing power of mobile GPUs. We evaluate PhoneBit with AlexNet, YOLOv2 Tiny and VGG16 with their binary version. Our experiment results show that PhoneBit can achieve significant speedup and energy efficiency compared with state-of-the-art frameworks for mobile devices.

cs.DC↗

arXiv2019

The role of initial speed in projectile impacts into light granular media

Projectile impact into a light granular material composed of expanded polypropylene (EPP) particles is investigated systematically with various impact velocities. Experimentally, the trajectory of an intruder moving inside the granular material is monitored with a recently developed non-invasive microwave radar system. Numerically, discrete element simulations together with coarse-graining techniques are employed to address both dynamics of the intruder and response of the granular bed. Our experimental and numerical results of the intruder dynamics agree with each other quantitatively and are in congruent with existing phenomenological model on granular drag. Stepping further, we explore the `microscopic' origin of granular drag through characterizing the response of granular bed, including density, velocity and kinetic stress fields at the mean-field level. In addition, we find that the dynamics of cavity collapse behind the intruder behaves qualitatively different with different impact velocities. Moreover, the kinetic pressure ahead of the intruder decays exponentially in the co-moving system of the intruder. Its scaling gives rise to a characteristic length scale, which is in the order of intruder size. This finding is in perfect agreement with the long-scale inertial dissipation type that we find in all cases.

cond-mat.soft↗

arXiv2019

BigDL: A Distributed Deep Learning Framework for Big Data

This paper presents BigDL (a distributed deep learning framework for Apache Spark), which has been used by a variety of users in the industry for building deep learning applications on production big data platforms. It allows deep learning applications to run on the Apache Hadoop/Spark cluster so as to directly process the production data, and as a part of the end-to-end data analysis pipeline for deployment and management. Unlike existing deep learning frameworks, BigDL implements distributed, data parallel training directly on top of the functional compute model (with copy-on-write and coarse-grained operations) of Spark. We also share real-world experience and "war stories" of users that have adopted BigDL to address their challenges(i.e., how to easily build end-to-end data analysis and deep learning pipelines for their production data).

cs.DC↗

arXiv2019

Correction of IQ mismatch for a particle tracking radar

For a better understanding of granular flow problems such as silo blockage, avalanche triggering, mixing and segregation, it is essential to have a `microscopic' view of individual particles. In order to cope with the difficulty arising from the opacity of granular materials, such as sands, powders and grains, a small scale bi-static radar system operating at $10$\,GHz (X-band) was recently introduced to trace a sub-centimeter particle in three dimensions. Similar to a moving target indicator radar, the relative movement of the tracer with respect to each of the three receiving antennae is obtained via comparing the phase shift of the electromagnetic wave traveling through the target area with an IQ-Mixer. From the azimuth and tilt angles of the receiving antennae obtained in the calibration, the target trajectory in a three-dimensional Cartesian system is reconstructed. Using a free-falling sphere as a test case, we discuss the accuracy of this radar system and possible ways to enhance it by IQ mismatch corrections.

physics.data-an↗

arXiv2019

A mathematical model of asynchronous data flow in parallel computers

We present a simplified model of data flow on processors in a high performance computing framework involving computations necessitating inter-processor communications. From this ordinary differential model, we take its asymptotic limit, resulting in a model which treats the computer as a continuum of processors and data flow as an Eulerian fluid governed by a conservation law. We derive a Hamilton-Jacobi equation associated with this conservation law for which the existence and uniqueness of solutions can be proven. We then present the results of numerical experiments for both discrete and continuum models; these show a qualitative agreement between the two and the effect of variations in the computing environment's processing capabilities on the progress of the modeled computation.

cs.DC↗

arXiv2019

First-principles prediction of a new family of layered topological insulators

Dozens of layered V$_2$IV$_2$VI$_6$ (V=P, As, Sb, Bi, IV=Si, Ge, Sn, Pb, VI=S, Se, Te) materials are investigated, several of which have been successfully synthesized in experiment. Among them, we predict nine strong topological insulators (TIs), two strong topological metals (TMs) and nearly twenty trivial insulators at their equilibrium structures. The TIs are in the (1;111) topological class, with energy gaps ranging from 0.04 to 0.2 eV. The strong TMs and the trivial insulators belong to the (1;111) and (0;000) topological classes, respectively. Small compressive strains easily turn some of the trivial insulators into strong TIs. This study enriches not only the family of topological materials but also the family of van der Waals layered materials, providing promising candidates for the future spintronic devices.

cond-mat.mtrl-sci↗

arXiv2019

Acoustics of Margravial Opera House Bayreuth

The Margravial Opera House Bayreuth, built between 1745 and 1750, is a well preserved Baroque court theatre designed by Giuseppe Galli Bibiena [1]. It provides an opportunity to experience not only the visual but also the acoustic design of opera theatres in the 18th century, as the bell-shaped auditorium along with the decoratively painted canvas remains intact. Using balloons and hand-claps as sound sources, we characterize the impulse response of this opera house after its recent renovation. The reverberation time (RT), early decay time (EDT) and clarity factor are characterized and discussed in comparison to historical Italian theatres of a similar age.

physics.pop-ph↗

arXiv2019

Energy-Efficient Slithering Gait Exploration for a Snake-like Robot based on Reinforcement Learning

Similar to their counterparts in nature, the flexible bodies of snake-like robots enhance their movement capability and adaptability in diverse environments. However, this flexibility corresponds to a complex control task involving highly redundant degrees of freedom, where traditional model-based methods usually fail to propel the robots energy-efficiently. In this work, we present a novel approach for designing an energy-efficient slithering gait for a snake-like robot using a model-free reinforcement learning (RL) algorithm. Specifically, we present an RL-based controller for generating locomotion gaits at a wide range of velocities, which is trained using the proximal policy optimization (PPO) algorithm. Meanwhile, a traditional parameterized gait controller is presented and the parameter sets are optimized using the grid search and Bayesian optimization algorithms for the purposes of reasonable comparisons. Based on the analysis of the simulation results, we demonstrate that this RL-based controller exhibits very natural and adaptive movements, which are also substantially more energy-efficient than the gaits generated by the parameterized controller. Videos are shown at \textcolor{blue}{\href{https://videoviewsite.wixsite.com/rlsnake}{https://videoviewsite.wixsite.com/rlsnake}}.

cs.RO↗

arXiv2019

3D Magnetic Reconnection with a spatially confined X-line extent -- Implications for Dipolarizing Flux Bundles and the Dawn-Dusk Asymmetry

Using 3D particle-in-cell (PIC) simulations, we study magnetic reconnection with the x-line being spatially confined in the current direction. We include thick current layers to prevent reconnection at two ends of a thin current sheet that has a thickness on an ion inertial (di) scale. The reconnection rate and outflow speed drop significantly when the extent of the thin current sheet in the current direction is < O(10 di). When the thin current sheet extent is long enough, we find it consists of two distinct regions; an inactive region (on the ion-drifting side) exists adjacent to the active region where reconnection proceeds normally as in a 2D case. The extent of this inactive region is ~ O(10 di), and it suppresses reconnection when the thin current sheet extent is comparable or shorter. The time-scale of current sheet thinning toward fast reconnection can be translated into the spatial-scale of this inactive region; because electron drifts inside the ion diffusion region transport the reconnected magnetic flux, that drives outflows and furthers the current sheet thinning, away from this region. This is a consequence of the Hall effect in 3D. While this inactive region may explain the shortest possible azimuthal extent of dipolarizing flux bundles at Earth, it may also explain the dawn-dusk asymmetry observed at the magnetotail of Mercury, that has a global dawn-dusk extent much shorter than that of Earth.

physics.plasm-ph↗