arXiv subjects

Fei Sun

Publications and source records attributed to Fei Sun.

At least 181 records · Page 10Linked to original sources

arXiv2019

Low-Power Computer Vision: Status, Challenges, Opportunities

Computer vision has achieved impressive progress in recent years. Meanwhile, mobile phones have become the primary computing platforms for millions of people. In addition to mobile phones, many autonomous systems rely on visual data for making decisions and some of these systems have limited energy (such as unmanned aerial vehicles also called drones and mobile robots). These systems rely on batteries and energy efficiency is critical. This article serves two main purposes: (1) Examine the state-of-the-art for low-power solutions to detect objects in images. Since 2015, the IEEE Annual International Low-Power Image Recognition Challenge (LPIRC) has been held to identify the most energy-efficient computer vision solutions. This article summarizes 2018 winners' solutions. (2) Suggest directions for research as well as opportunities for low-power computer vision.

cs.CV↗

arXiv2019

Value-aware Recommendation based on Reinforced Profit Maximization in E-commerce Systems

Existing recommendation algorithms mostly focus on optimizing traditional recommendation measures, such as the accuracy of rating prediction in terms of RMSE or the quality of top-$k$ recommendation lists in terms of precision, recall, MAP, etc. However, an important expectation for commercial recommendation systems is to improve the final revenue/profit of the system. Traditional recommendation targets such as rating prediction and top-$k$ recommendation are not directly related to this goal. In this work, we blend the fundamental concepts in online advertising and micro-economics into personalized recommendation for profit maximization. Specifically, we propose value-aware recommendation based on reinforcement learning, which directly optimizes the economic value of candidate items to generate the recommendation list. In particular, we generalize the basic concept of click conversion rate (CVR) in computational advertising into the conversation rate of an arbitrary user action (XVR) in E-commerce, where the user actions can be clicking, adding to cart, adding to wishlist, etc. In this way, each type of user action is mapped to its monetized economic value. Economic values of different user actions are further integrated as the reward of a ranking list, and reinforcement learning is used to optimize the recommendation list for the maximum total value. Experimental results in both offline benchmarks and online commercial systems verified the improved performance of our framework, in terms of both traditional top-$k$ ranking tasks and the economic profits of the system.

cs.IR↗

arXiv2018

ChamNet: Towards Efficient Network Design through Platform-Aware Model Adaptation

This paper proposes an efficient neural network (NN) architecture design methodology called Chameleon that honors given resource constraints. Instead of developing new building blocks or using computationally-intensive reinforcement learning algorithms, our approach leverages existing efficient network building blocks and focuses on exploiting hardware traits and adapting computation resources to fit target latency and/or energy constraints. We formulate platform-aware NN architecture search in an optimization framework and propose a novel algorithm to search for optimal architectures aided by efficient accuracy and resource (latency and/or energy) predictors. At the core of our algorithm lies an accuracy predictor built atop Gaussian Process with Bayesian optimization for iterative sampling. With a one-time building cost for the predictors, our algorithm produces state-of-the-art model architectures on different platforms under given constraints in just minutes. Our results show that adapting computation resources to building blocks is critical to model performance. Without the addition of any bells and whistles, our models achieve significant accuracy improvements against state-of-the-art hand-crafted and automatically designed architectures. We achieve 73.8% and 75.3% top-1 accuracy on ImageNet at 20ms latency on a mobile CPU and DSP. At reduced latency, our models achieve up to 8.5% (4.8%) and 6.6% (9.3%) absolute top-1 accuracy improvements compared to MobileNetV2 and MnasNet, respectively, on a mobile CPU (DSP), and 2.7% (4.6%) and 5.6% (2.6%) accuracy gains over ResNet-101 and ResNet-152, respectively, on an Nvidia GPU (Intel CPU).

cs.CV↗

arXiv2018

Multi-Source Pointer Network for Product Title Summarization

In this paper, we study the product title summarization problem in E-commerce applications for display on mobile devices. Comparing with conventional sentence summarization, product title summarization has some extra and essential constraints. For example, factual errors or loss of the key information are intolerable for E-commerce applications. Therefore, we abstract two more constraints for product title summarization: (i) do not introduce irrelevant information; (ii) retain the key information (e.g., brand name and commodity name). To address these issues, we propose a novel multi-source pointer network by adding a new knowledge encoder for pointer network. The first constraint is handled by pointer mechanism. For the second constraint, we restore the key information by copying words from the knowledge encoder with the help of the soft gating mechanism. For evaluation, we build a large collection of real-world product titles along with human-written short titles. Experimental results demonstrate that our model significantly outperforms the other baselines. Finally, online deployment of our proposed model has yielded a significant business impact, as measured by the click-through rate.

cs.CL↗

arXiv2018

2018 Low-Power Image Recognition Challenge

The Low-Power Image Recognition Challenge (LPIRC, https://rebootingcomputing.ieee.org/lpirc) is an annual competition started in 2015. The competition identifies the best technologies that can classify and detect objects in images efficiently (short execution time and low energy consumption) and accurately (high precision). Over the four years, the winners' scores have improved more than 24 times. As computer vision is widely used in many battery-powered systems (such as drones and mobile phones), the need for low-power computer vision will become increasingly important. This paper summarizes LPIRC 2018 by describing the three different tracks and the winners' solutions.

cs.CV↗

arXiv2018

Estimation of the chiral magnetic effect considering the magnetic field response of the QGP medium

The magnetic field plays a major role in the searching of the chiral magnetic effect in relativistic heavy-ion collisions. If the lifetime of the magnetic field is too short, as expected by simulations of the field in the vacuum, the chiral magnetic effect will be largely suppressed. However, the lifetime of the magnetic field will become longer when the QGP medium response is considered. We give an estimate of the effect, especially considering the magnetic field response of the QGP medium, and compare it with the experimental results of background-subtracted correlator $H$ at RHIC and LHC energies. The results show that our method explains better for the experimental results at the top RHIC energy than that of the LHC energy.

nucl-th↗

arXiv2017

An acoustic metamaterial lens for acoustic point-to-point communication in air

Acoustic metamaterials have become a novel and effective way to control sound waves and design acoustic devices. In this study, we design a 3D acoustic metamaterial lens (AML) to achieve point-to-point acoustic communication in air: any acoustic source (i.e. a speaker) in air enclosed by such an AML can produce an acoustic image where the acoustic wave is focused (i.e. the field intensity is at a maximum, and the listener can receive the information), while the acoustic field at other spatial positions is low enough that listeners can hear almost nothing. Unlike a conventional elliptical reflective mirror, the acoustic source can be moved around inside our proposed AML. Numerical simulations are given to verify the performance of the proposed AML.

physics.gen-ph↗

arXiv2017

125 GeV Higgs decay with lepton flavor violation in the $μν$SSM

Recently, the CMS and ATLAS Collaborations have reported direct searches for the 125 GeV Higgs decay with lepton flavor violation, $h\rightarrow μτ$. In this work, we analyze the signal of the lepton flavour violating (LFV) Higgs decay $h\rightarrow μτ$ in the $μ$ from $ν$ Supersymmetric Standard Model ($μν$SSM) with slepton flavor mixing. Simultaneously, we consider the constraints from the LFV decay $τ\rightarrow μγ$, the muon anomalous magnetic dipole moment and the lightest Higgs mass around 125 GeV.

hep-ph↗

arXiv2016

Increasing efficiency of a wireless energy transfer system by spatial translational transformation

An optical translational projector (OTP) designed by transformation optics is applied to improve the energy transfer efficiency in a wireless energy transfer (WET) system. Numerical simulation results show our OTP can greatly enhance the energy transfer efficiency (e.g. nearly 2 orders, compared to the case without our OTP) in WET systems, which is much larger than previous methods (e.g. magnetic super-lens). A 3D reduced OTP composed by layered isotropic magnetic materials is designed, whose performance has been verified by 3D numerical simulations in 10MHz. We also study the influence of loss of metamaterials on the performance of proposed OTP.

physics.gen-ph↗

arXiv2016

Computations of superstring amplitudes in pure spinor formalism via Cadabra

The discovery of pure spinor formalism makes the computation of superstring scattering amplitudes possible. In this paper, we will illustrate how computer algebra system Cadabra is used in computing the supersymmetric amplitude in pure spinor formalism and provide the source code that computes the tree-level massless 5-gluon amplitude.

hep-th↗

arXiv2016

Web Based Teleoperation of a Humanoid Robot

The Cloud-based Advanced Robotics Laboratory (CARL) integrates a whole body controller and web-based teleoperation to enable any device with a web browser to access and control a humanoid robot. By integrating humanoid robots with the cloud, they are accessible from any Internet-connected device. Increased accessibility is important because few people have access to state-of-the-art humanoid robots limiting their rate of development. CARL's implementation is based on modern software libraries, frameworks, and middleware including Node.js, Socket.IO, ZMQ, ROS, Robot Web Tools, and ControlIt! Feasibility is demonstrated by having inexperienced human operators use a smartphone's web-browser to control Dreamer, a torque-controlled humanoid robot based on series elastic actuators, and make it perform a dual-arm manipulation task. The implementation serves as a proof-of-concept and foundation upon which many advanced humanoid robot technologies can be researched and developed.

cs.RO↗

arXiv2016

Estimation of CD4+ T Cell Count Parameters in HIV/AIDS Patients Based on Real-time Nonlinear Receding Horizon Control

An increasing number of control techniques are introduced to HIV infection problem to explore the options of helping clinical testing, optimizing drug treatments and to study the drug resistance situations. In such cases, complete/accurate knowledge of the HIV model and/or parameters is critical not only to monitor the dynamics of the system, but also to adjust the therapy accordingly. In those studies, existence of any type of unknown parameters imposes severe set-backs and becomes problematic for the treatment of the patients. In this work, we develop a real-time adaptive nonlinear receding horizon control approach to aid such scenarios and to estimate unknown constant/time-varying parameters of nonlinear HIV system models. For this purpose, the problem of estimation is updated by a series of finite-time optimization problem which can be solved by backwards sweep Riccati method in real time without employing any iteration techniques. The simulation results demonstrates the fact that proposed algorithm is able to estimate unknown constant/time-varying parameters of HIV/AIDS model effectively and provide a unique, adaptive solution methodology to an important open problem.

math.OC↗

arXiv2016

Algebraic K-Theory and Modular Symbols

In this paper, we calculate the differential $d^1$ of the rank spectral sequence. We generalize Quillen's spectral sequence from Dedekind domain to general integral Noetherian ring $A$ by considering the Q-construction $Q^{tf}(A)$ of the category of finitely generated torsion-free modules. In particular, by resolution theorem, if $A$ is regular, then $Q^{tf}(A)$ is homotopy equivalent to $QP(A)$ where $P(A)$ is the category of finitely generated projective $A$-modules. We deduce the differential $d^1$ by using Tits buildings, Steinberg modules, and modular symbols in the sense of Ash-Rudolph. The spirit of Quillen's categorical homotopy theory will be used intensively throughout this paper.

math.AT↗

arXiv2016

Semantic Regularities in Document Representations

Recent work exhibited that distributed word representations are good at capturing linguistic regularities in language. This allows vector-oriented reasoning based on simple linear algebra between words. Since many different methods have been proposed for learning document representations, it is natural to ask whether there is also linear structure in these learned representations to allow similar reasoning at document level. To answer this question, we design a new document analogy task for testing the semantic regularities in document representations, and conduct empirical evaluations over several state-of-the-art document representation models. The results reveal that neural embedding based document representations work better on this analogy task than conventional methods, and we provide some preliminary explanations over these observations.

cs.CL↗

arXiv2016

Nonlinear Consensus Strategies for Multi-Agent Networks in Presence of Communication Delays and Switching Topologies: Real-Time Receding Horizon Approach

This paper presents a novel framework which combines a non-iterative solution of Real-Time Nonlinear Receding Horizon Control (NRHC) methodology to achieve consensus within complex network topologies with existing time-delays and in presence of switching topologies. In this formulation, we solve the distributed nonlinear optimization problem for multi-agent network systems directly, \emph{in real-time}, without any dependency on iterative processes, where the stability and convergence guarantees are provided for the solution. Three benchmark examples on non-linear chaotic systems provide validated results which demonstrate the significant outcomes of such methodology.

math.OC↗

arXiv2015

Distributed Real-Time Non-Linear Receding Horizon Control Methodology for Multi-Agent Consensus Problems

This work investigates the consensus problem for multi-agent nonlinear systems through the distributed real-time nonlinear receding horizon control methodology. With this work, we develop a scheme to reach the consensus for nonlinear multi agent systems under fixed directed/undirected graph(s) without the need of any linearization techniques. For this purpose, the problem of consensus is converted into an optimization problem and is directly solved by the backwards sweep Riccati method to generate the control protocol which results in a non-iterative algorithm. Stability analysis is conducted to provide convergence guarantees of proposed scheme. In addition, an extension to the leader-following consensus of nonlinear multi-agent systems is presented. Several examples are provided to validate and demonstrate the effectiveness of the presented scheme and the corresponding theoretical results.

math.OC↗

arXiv2015

Component-Enhanced Chinese Character Embeddings

Distributed word representations are very useful for capturing semantic information and have been successfully applied in a variety of NLP tasks, especially on English. In this work, we innovatively develop two component-enhanced Chinese character embedding models and their bigram extensions. Distinguished from English word embeddings, our models explore the compositions of Chinese characters, which often serve as semantic indictors inherently. The evaluations on both word similarity and text classification demonstrate the effectiveness of our models.

cs.CL↗

arXiv2015

Real-Time Non-Linear Receding Horizon Control Methodology for Estimation of Time-Varying Parameters

In control and engineering community, models generally contain a number of parameters which are unknown or roughly known. A complete knowledge of these parameters is critical to describe and analyze the dynamics of the system. This paper develops a novel approach of estimating unknown time-varying parameters of nonlinear systems based on real-time nonlinear receding horizon control strategy. For this purpose, the problem of estimation is converted into optimization problem which can be solved by backwards sweep Riccati method. Methodology is implemented on two nonlinear examples without providing any reference of the parameters. Results successfully demonstrate the fact that proposed algorithm is able to estimate unknown time-varying parameters of nonlinear systems effectively.

math.OC↗