arXiv subjects

Yu Tang

Publications and source records attributed to Yu Tang.

51 records · Page 3Linked to original sources

arXiv2022

DELTA: Dynamically Optimizing GPU Memory beyond Tensor Recomputation

The further development of deep neural networks is hampered by the limited GPU memory resource. Therefore, the optimization of GPU memory resources is highly demanded. Swapping and recomputation are commonly applied to make better use of GPU memory in deep learning. However, as an emerging domain, several challenges remain:1)The efficiency of recomputation is limited for both static and dynamic methods. 2)Swapping requires offloading parameters manually, which incurs a great time cost. 3) There is no such dynamic and fine-grained method that involves tensor swapping together with tensor recomputation nowadays. To remedy the above issues, we propose a novel scheduler manager named DELTA(Dynamic tEnsor offLoad and recompuTAtion). To the best of our knowledge, we are the first to make a reasonable dynamic runtime scheduler on the combination of tensor swapping and tensor recomputation without user oversight. In DELTA, we propose a filter algorithm to select the optimal tensors to be released out of GPU memory and present a director algorithm to select a proper action for each of these tensors. Furthermore, prefetching and overlapping are deliberately considered to overcome the time cost caused by swapping and recomputing tensors. Experimental results show that DELTA not only saves 40%-70% of GPU memory, surpassing the state-of-the-art method to a great extent but also gets comparable convergence results as the baseline with acceptable time delay. Also, DELTA gains 2.04$\times$ maximum batchsize when training ResNet-50 and 2.25$\times$ when training ResNet-101 compared with the baseline. Besides, comparisons between the swapping cost and recomputation cost in our experiments demonstrate the importance of making a reasonable dynamic scheduler on tensor swapping and tensor recomputation, which refutes the arguments in some related work that swapping should be the first and best choice.

cs.LG↗

arXiv2022

Single Molecule Mixture: A Concept in Polymer Science

In theory, there exist two extreme forms of substances: pure form and single-molecule mixture form. Single-molecule mixture form contains a mixture of molecules that have molecularly different structures. This elusive form has not yet been explored. Herein, we report a study of single molecule mixture state by a combination of model construction and mathematical analysis, and a series of interesting results were obtained. These results provide theoretical evidence that single-molecule mixture state may indeed exist in realistic synthetic or natural polymer system.

cond-mat.mtrl-sci↗

arXiv2022

SAD: A Large-scale Dataset towards Airport Detection in Synthetic Aperture Radar Images

Airports have an important role in both military and civilian domains. The synthetic aperture radar (SAR) based airport detection has received increasing attention in recent years. However, due to the high cost of SAR imaging and annotation process, there is no publicly available SAR dataset for airport detection. As a result, deep learning methods have not been fully used in airport detection tasks. To provide a benchmark for airport detection research in SAR images, this paper introduces a large-scale SAR Airport Dataset (SAD). In order to adequately reflect the demands of real world applications, it contains 624 SAR images from Sentinel 1B and covers 104 airfield instances with different scales, orientations and shapes. The experiments of multiple deep learning approach on this dataset proves its effectiveness. It developing state-of-the-art airport area detection algorithms or other relevant tasks.

cs.CV↗

arXiv2022

P-ADMMiRNN: Training RNN with Stable Convergence via An Efficient and Paralleled ADMM Approach

It is hard to train Recurrent Neural Network (RNN) with stable convergence and avoid gradient vanishing and exploding problems, as the weights in the recurrent unit are repeated from iteration to iteration. Moreover, RNN is sensitive to the initialization of weights and bias, which brings difficulties in training. The Alternating Direction Method of Multipliers (ADMM) has become a promising algorithm to train neural networks beyond traditional stochastic gradient algorithms with the gradient-free features and immunity to unsatisfactory conditions. However, ADMM could not be applied to train RNN directly since the state in the recurrent unit is repetitively updated over timesteps. Therefore, this work builds a new framework named ADMMiRNN upon the unfolded form of RNN to address the above challenges simultaneously. We also provide novel update rules and theoretical convergence analysis. We explicitly specify essential update rules in the iterations of ADMMiRNN with constructed approximation techniques and solutions to each sub-problem instead of vanilla ADMM. Numerical experiments are conducted on MNIST, IMDb, and text classification tasks. ADMMiRNN achieves convergent results and outperforms the compared baselines. Furthermore, ADMMiRNN trains RNN more stably without gradient vanishing or exploding than stochastic gradient algorithms. We also provide a distributed paralleled algorithm regarding ADMMiRNN, named P-ADMMiRNN, including Synchronous Parallel ADMMiRNN (SP-ADMMiRNN) and Asynchronous Parallel ADMMiRNN (AP-ADMMiRNN), which is the first to train RNN with ADMM in an asynchronous parallel manner. The source code is publicly available.

cs.LG↗

arXiv2022

Resilient Ramp Control for Highways Facing Stochastic Perturbations

Highway capacity is often subject to stochastic perturbations due to the combined effects of weather, traffic mixture, driver behavior, etc. This paper is motivated by the need of a systematic approach to traffic control with performance guarantees in the face of such perturbations. We develop a novel control-theoretic method for designing perturbation-resilient ramp metering. We consider a cell-transmission model with 1) Markovian cell capacities and 2) buffers representing on-ramps and upstream mainline. Using this model, we analyze the stability of on-ramp queues by constructing piecewise Lyapunov functions that consider the nature of nonlinear traffic dynamics. Then, we design ramp controllers that guarantee bounds for throughput and queue sizes. We also formulate the problem of coordinated ramp metering as a bi-level optimization with non-convex inner sub-problems. To address the computational issue in solving this problem, we also consider localized and partially coordinated reformulations. A case study of a 18.1-km highway in Los Angeles, USA indicates a 8.3\% (resp. 9.9\%) reduction of vehicle-hours-traveled obtained by the localized (resp. partially coordinated) control, both outperforming the classical ALINEA and METALINE controllers.

math.OC↗

arXiv2022

A Novel Four-DOF Lagrangian Approach to Attitude Tracking for Rigid Spacecraft

This paper presents a novel Lagrangian approach to attitude tracking for rigid spacecraft using unit quaternions, where the motion equations of a spacecraft are described by a four degrees of freedom Lagrangian dynamics subject to a holonomic constraint imposed by the norm of a unit quaternion. The basic energy-conservation property as well as some additional useful properties of the Lagrangian dynamics are explored, enabling to develop quaternion-based attitude tracking controllers by taking full advantage of a broad class of tracking control designs for mechanical systems based on energy-shaping methodology. Global tracking of a desired attitude on the unit sphere is achieved by designing control laws that render the tracking error on the four-dimensional Euclidean space to converge to the origin. The topological constraints for globally exponentially tracking by a quaternion-based continuous controller and singularities in controller designs based on any three-parameter representation of the attitude are then avoided. Using this approach, a full-state feedback controller is first developed, and then several important issues, such as robustness to noise in quaternion measurements, unknown on-orbit torque disturbances, uncertainty in the inertial matrix, and lack of angular-velocity measurements are addressed progressively, by designing a hybrid state-feedback controller, an adaptive hybrid state-feedback controller, and an adaptive hybrid attitude-feedback controller. Global asymptotic stability is established for each controller. Simulations are included to illustrate the theoretical results.

eess.SY↗

arXiv2021

Thermoelectric and stress distributions around a smooth cavity in thermoelectric material

Thermoelectric materials have attracted more and more attention since they are friendly to the environment and have potentials for sustainable and renewable energy applications. As typically brittle semiconductors with low mechanical strength and always subjected to defects and damages, to clarify the stress concentration is very important in the design and implement of thermoelectric devices. The two-dimensional thermoelectric coupling problem due to a cavity embedded in an infinite isotropic homogeneous thermoelectric material, subjected to uniform electric current density or uniform energy flux, is studied, where the shape of the cavity is characterized by the Laurent polynomial, and the electric insulated and adiabatic boundary around the cavity are considered. The explicit analytic solutions of Kolosov-Muskhelishvili (K-M) potentials and rigid-body translation are carried out through a novel tactic. Comparing with the reported results, the new obtained are completely exact and possess a finite form. Some results of three typical cavities are presented to analyze the electric current densities (energy fluxes) and stresses around the tips. The main conclusions include: the distribution of thermoelectric field and stress at the tip obviously depends on the curvature of the contour and loading directions; for triangle and square with symmetrical tips, the maximum thermoelectric and stress concentration reach the maximum or minimum when the loading direction is parallel to or perpendicular to the symmetry axis of the tip, which is distinct to the extremum characteristics of pentagram with bimodal of curvature around the tip; the maximum thermoelectric and stress concentration appear near the maximum curvature point for most load directions, but not at the maximum curvature point.

physics.app-ph↗

arXiv2021

Thermal stress around a smooth cavity in a plate subjected to uniform heat flux

The two-dimensional thermoelastic problem of an adiabatic cavity in an infinite isotropic homogeneous medium subjected to uniform heat flux is studied, where the shape of the cavity is characterized by the Laurent polynomial. By virtue of a novel tactics, the obtained K-M potentials can be explicitly worked out to satisfy the boundary conditions precisely, and the possible translation of the cavity is also available. The new and explicit analytical solutions are compared with the those reported in literature and some serious problems are found and corrected. Finally, some discussions on the thermal stress concentration around the tips of three typical cavities are provided.

physics.app-ph↗

arXiv2020

A Deep Reinforcement Learning Approach for Ramp Metering Based on Traffic Video Data

Ramp metering that uses traffic signals to regulate vehicle flows from the on-ramps has been widely implemented to improve vehicle mobility of the freeway. Previous studies generally update signal timings in real-time based on predefined traffic measures collected by point detectors, such as traffic volumes and occupancies. Comparing with point detectors, traffic cameras-which have been increasingly deployed on road networks-could cover larger areas and provide more detailed traffic information. In this work, we propose a deep reinforcement learning (DRL) method to explore the potential of traffic video data in improving the efficiency of ramp metering. The proposed method uses traffic video frames as inputs and learns the optimal control strategies directly from the high-dimensional visual inputs. A real-world case study demonstrates that, in comparison with a state-of-the-practice method, the proposed DRL method results in 1) lower travel times in the mainline, 2) shorter vehicle queues at the on-ramp, and 3) higher traffic flows downstream of the merging area. The results suggest that the proposed method is able to extract useful information from the video data for better ramp metering controls.

cs.CV↗

arXiv2020

Median regression with differential privacy

Median regression analysis has robustness properties which make it attractive compared with regression based on the mean, while differential privacy can protect individual privacy during statistical analysis of certain datasets. In this paper, three privacy preserving methods are proposed for median regression. The first algorithm is based on a finite smoothing method, the second provides an iterative way and the last one further employs the greedy coordinate descent approach. Privacy preserving properties of these three methods are all proved. Accuracy bound or convergence properties of these algorithms are also provided. Numerical calculation shows that the first method has better accuracy than the others when the sample size is small. When the sample size becomes larger, the first method needs more time while the second method needs less time with well-matched accuracy. For the third method, it costs less time in both cases, while it highly depends on step size.

stat.CO↗

arXiv2019

Security Risk Analysis of the Shorter-Queue Routing Policy for Two Symmetric Servers

In this article, we study the classical shortest queue problem under the influence of malicious attacks, which is relevant to a variety of engineering system including transportation, manufacturing, and communications. We consider a homogeneous Poisson arrival process of jobs and two parallel exponential servers with symmetric service rates. A system operator route incoming jobs to the shorter queue; if the queues are equal, the job is routed randomly. A malicious attacker is able to intercept the operator's routing instruction and overwrite it with a randomly generated one. The operator is able to defend individual jobs to ensure correct routing. Both attacking and defending induce technological costs. The attacker's (resp. operator's) decision is the probability of attacking (resp. defending) the routing of each job. We first quantify the queuing cost for given strategy profiles by deriving a theoretical upper bound for the cost. Then, we formulate a non-zero-sum attacker-defender game, characterize the equilibria in multiple regimes, and quantify the security risk. We find that the attacker's best strategy is either to attack all jobs or not to attack, and the defender's strategy is strongly influenced by the arrival rate of jobs. Finally, as a benchmark, we compare the security risks of the feedback-controlled system to a corresponding open-loop system with Bernoulli routing. We show that the shorter-queue policy has a higher (resp. lower) security risk than the Bernoulli policy if the demand is lower (resp. higher) than the service rate of one server.

eess.SY↗

arXiv2015

Vibration Control Design for Nonlinear Systems using Frequency Response Function

A nonlinear frequency response based adaptive vibration controller is proposed for a class of nonlinear mechanical systems. In order to obtain the nonlinear Frequency Response Function (FRF), the convergence properties of the system are studied by using the convergence (contraction) theory. If the system under consideration is: 1) convergent, it directly enables to derive a nonlinear FRF for a band of excitation inputs, 2) non-convergent, first a controller is used to obtain the convergence and then the corresponding FRF for a band of excitation inputs is derived. Now the gains of the proposed adaptive controller are tuned such that a desired closed-loop frequency response, in the presence of excitation inputs is achieved. Finally, a building structure with nonlinear cubic stiffness and a satellite system are considered to illustrate the theoretical results.

eess.SY↗

arXiv2014

Optimum mixed level detecting arrays

As a type of search design, a detecting array can be used to generate test suites to identify and detect faults caused by interactions of factors in a component-based system. Recently, the construction and optimality of detecting arrays have been investigated in depth in the case where all the factors are assumed to have the same number of levels. However, for real world applications, it is more desirable to use detecting arrays in which the various factors may have different numbers of levels. This paper gives a general criterion to measure the optimality of a mixed level detecting array in terms of its size. Based on this optimality criterion, the combinatorial characteristics of mixed level detecting arrays of optimum size are investigated. This enables us to construct optimum mixed level detecting arrays with a heuristic optimization algorithm and combinatorial methods. As a result, some existence results for optimum mixed level detecting arrays achieving a lower bound are provided for practical use.

math.ST↗

arXiv2012

Uniform fractional factorial designs

The minimum aberration criterion has been frequently used in the selection of fractional factorial designs with nominal factors. For designs with quantitative factors, however, level permutation of factors could alter their geometrical structures and statistical properties. In this paper uniformity is used to further distinguish fractional factorial designs, besides the minimum aberration criterion. We show that minimum aberration designs have low discrepancies on average. An efficient method for constructing uniform minimum aberration designs is proposed and optimal designs with 27 and 81 runs are obtained for practical use. These designs have good uniformity and are effective for studying quantitative factors.

math.ST↗

arXiv1994

Capillary Rise in Tubes with Sharp Grooves

Liquid in grooved capillaries, made by e.g. inserting a plate in a cylindrical tube, exhibits unusual spreading and flow properties. One example is capillary rise, where a long, upward tongue on top of the usual meniscus has been observed along the groove. We attribute the underlying mechanism to a thermodynamic instability against spreading for a (partial or complete wetting) liquid in a sharp groove whose opening angle $α$ is less than a critical value $α_c=π-2θ$. The equilibrium shape of the tongue is determined analytically. The dynamics of liquid rising is studied in the viscous regime. When the diameter of the tube is smaller than the capillary length, the center part of the meniscus rises with time $t$ following a $t^{1/2}$-law, while the tongue is truncated at a height which grows following a $t^{1/3}$-law. Sharp groove also facilitates release of gas bubbles trapped inside a capillary under the action of gravity.

cond-mat↗