arXiv subjects

Meng Li

Publications and source records attributed to Meng Li.

At least 343 records · Page 19Linked to original sources

arXiv2020

SID: Incremental Learning for Anchor-Free Object Detection via Selective and Inter-Related Distillation

Incremental learning requires a model to continually learn new tasks from streaming data. However, traditional fine-tuning of a well-trained deep neural network on a new task will dramatically degrade performance on the old task -- a problem known as catastrophic forgetting. In this paper, we address this issue in the context of anchor-free object detection, which is a new trend in computer vision as it is simple, fast, and flexible. Simply adapting current incremental learning strategies fails on these anchor-free detectors due to lack of consideration of their specific model structures. To deal with the challenges of incremental learning on anchor-free object detectors, we propose a novel incremental learning paradigm called Selective and Inter-related Distillation (SID). In addition, a novel evaluation metric is proposed to better assess the performance of detectors under incremental learning conditions. By selective distilling at the proper locations and further transferring additional instance relation knowledge, our method demonstrates significant advantages on the benchmark datasets PASCAL VOC and COCO.

cs.CV↗

arXiv2020

Bayesian Median Autoregression for Robust Time Series Forecasting

We develop a Bayesian median autoregressive (BayesMAR) model for time series forecasting. The proposed method utilizes time-varying quantile regression at the median, favorably inheriting the robustness of median regression in contrast to the widely used mean-based methods. Motivated by a working Laplace likelihood approach in Bayesian quantile regression, BayesMAR adopts a parametric model bearing the same structure as autoregressive models by altering the Gaussian error to Laplace, leading to a simple, robust, and interpretable modeling strategy for time series forecasting. We estimate model parameters by Markov chain Monte Carlo. Bayesian model averaging is used to account for model uncertainty, including the uncertainty in the autoregressive order, in addition to a Bayesian model selection approach. The proposed methods are illustrated using simulations and real data applications. An application to U.S. macroeconomic data forecasting shows that BayesMAR leads to favorable and often superior predictive performance compared to the selected mean-based alternatives under various loss functions that encompass both point and probabilistic forecasts. The proposed methods are generic and can be used to complement a rich class of methods that build on autoregressive models.

stat.AP↗

arXiv2020

ScaleNAS: One-Shot Learning of Scale-Aware Representations for Visual Recognition

Scale variance among different sizes of body parts and objects is a challenging problem for visual recognition tasks. Existing works usually design dedicated backbone or apply Neural architecture Search(NAS) for each task to tackle this challenge. However, existing works impose significant limitations on the design or search space. To solve these problems, we present ScaleNAS, a one-shot learning method for exploring scale-aware representations. ScaleNAS solves multiple tasks at a time by searching multi-scale feature aggregation. ScaleNAS adopts a flexible search space that allows an arbitrary number of blocks and cross-scale feature fusions. To cope with the high search cost incurred by the flexible space, ScaleNAS employs one-shot learning for multi-scale supernet driven by grouped sampling and evolutionary search. Without further retraining, ScaleNet can be directly deployed for different visual recognition tasks with superior performance. We use ScaleNAS to create high-resolution models for two different tasks, ScaleNet-P for human pose estimation and ScaleNet-S for semantic segmentation. ScaleNet-P and ScaleNet-S outperform existing manually crafted and NAS-based methods in both tasks. When applying ScaleNet-P to bottom-up human pose estimation, it surpasses the state-of-the-art HigherHRNet. In particular, ScaleNet-P4 achieves 71.6% AP on COCO test-dev, achieving new state-of-the-art result.

cs.CV↗

arXiv2020

Equivalence of Convergence Rates of Posterior Distributions and Bayes Estimators for Functions and Nonparametric Functionals

We study the posterior contraction rates of a Bayesian method with Gaussian process priors in nonparametric regression and its plug-in property for differential operators. For a general class of kernels, we establish convergence rates of the posterior measure of the regression function and its derivatives, which are both minimax optimal up to a logarithmic factor for functions in certain classes. Our calculation shows that the rate-optimal estimation of the regression function and its derivatives share the same choice of hyperparameter, indicating that the Bayes procedure remarkably adapts to the order of derivatives and enjoys a generalized plug-in property that extends real-valued functionals to function-valued functionals. This leads to a practically simple method for estimating the regression function and its derivatives, whose finite sample performance is assessed using simulations. Our proof shows that, under certain conditions, to any convergence rate of Bayes estimators there corresponds the same convergence rate of the posterior distributions (i.e., posterior contraction rate), and vice versa. This equivalence holds for a general class of Gaussian processes and covers the regression function and its derivative functionals, under both the $L_2$ and $L_{\infty}$ norms. In addition to connecting these two fundamental large sample properties in Bayesian and non-Bayesian regimes, such equivalence enables a new routine to establish posterior contraction rates by calculating convergence rates of nonparametric point estimators. At the core of our argument is an operator-theoretic framework for kernel ridge regression and equivalent kernel techniques. We derive a range of sharp non-asymptotic bounds that are pivotal in establishing convergence rates of nonparametric point estimators and the equivalence theory, which may be of independent interest.

math.ST↗

arXiv2020

KeepAugment: A Simple Information-Preserving Data Augmentation Approach

Data augmentation (DA) is an essential technique for training state-of-the-art deep learning systems. In this paper, we empirically show data augmentation might introduce noisy augmented examples and consequently hurt the performance on unaugmented data during inference. To alleviate this issue, we propose a simple yet highly effective approach, dubbed \emph{KeepAugment}, to increase augmented images fidelity. The idea is first to use the saliency map to detect important regions on the original images and then preserve these informative regions during augmentation. This information-preserving strategy allows us to generate more faithful training examples. Empirically, we demonstrate our method significantly improves on a number of prior art data augmentation schemes, e.g. AutoAugment, Cutout, random erasing, achieving promising results on image classification, semi-supervised image classification, multi-view multi-camera tracking and object detection.

cs.CV↗

arXiv2020

Efficient in-situ image and video compression through probabilistic image representation

Fast and effective image compression for multi-dimensional images has become increasingly important for efficient storage and transfer of massive amounts of high-resolution images and videos. Desirable properties in compression methods include (1) high reconstruction quality at a wide range of compression rates while preserving key local details, (2) computational scalability, (3) applicability to a variety of different image/video types and of different dimensions, (4) progressive transmission, and (5) ease of tuning. We present such a method for multi-dimensional image compression called Compression via Adaptive Recursive Partitioning (CARP). CARP uses an optimal permutation of the image pixels inferred from a Bayesian probabilistic model on recursive partitions of the image to reduce its effective dimensionality, achieving a parsimonious representation that preserves information. CARP uses a multi-layer Bayesian hierarchical model to achieve in-situ compression along with self-tuning and regularization, with just one single parameter to be specified by the user to achieve the desired compression rate. Extensive numerical experiments using a variety of datasets including 2D still images, real-life YouTube videos, and surveillance videos show that CARP dominates the state-of-the-art image/video compression approaches---including JPEG, JPEG2000, BPG, MPEG4, HEVC and a neural network-based method---for all of these different image types and on nearly all of the individual images and videos over some methods.

eess.IV↗

arXiv2020

Learning Asymmetric and Local Features in Multi-Dimensional Data through Wavelets with Recursive Partitioning

Effective learning of asymmetric and local features in images and other data observed on multi-dimensional grids is a challenging objective critical for a wide range of image processing applications involving biomedical and natural images. It requires methods that are sensitive to local details while fast enough to handle massive numbers of images of ever increasing sizes. We introduce a probabilistic model-based framework that achieves these objectives by incorporating adaptivity into discrete wavelet transforms (DWT) through Bayesian hierarchical modeling, thereby allowing wavelet bases to adapt to the geometric structure of the data while maintaining the high computational scalability of wavelet methods---linear in the sample size (e.g., the resolution of an image). We derive a recursive representation of the Bayesian posterior model which leads to an exact message passing algorithm to complete learning and inference. While our framework is applicable to a range of problems including multi-dimensional signal processing, compression, and structural learning, we illustrate its work and evaluate its performance in the context of image reconstruction using real images from the ImageNet database, two widely used benchmark datasets, and a dataset from retinal optical coherence tomography and compare its performance to state-of-the-art methods based on basis transforms and deep learning.

stat.ME↗

arXiv2020

Entangled state generation via quantum walks with multiple coins

Generation of entangled state is of paramount importance both from quantum theoretical foundation and technology applications. Entanglement swapping provides an efficient method to generate entanglement in quantum communication protocols. However, perfect Bell measurements for qudits, the key to entanglement swapping, have been proven impossible to achieve by using only linear elements and particle detectors. To avoid this bottleneck, we propose a novel scheme to generate entangled state including two-qubit entangled state, two-qudit entangled state, three-qubit GHZ state and three-qudit GHZ state between several designate parties via the model of quantum walks with multiple coins. Then we conduct experimental realization of Bell state and three-qubit GHZ state between several designate parties on IBM quantum platform and the result has high fidelity by preforming quantum tomography. In the end, we give a practical application of our scheme in multiparty quantum secret sharing.

quant-ph↗

arXiv2020

An efficient second-order energy stable BDF scheme for the space fractional Cahn-Hilliard equation

The space fractional Cahn-Hilliard phase-field model is more adequate and accurate in the description of the formation and phase change mechanism than the classical Cahn-Hilliard model. In this article, we propose a temporal second-order energy stable scheme for the space fractional Cahn-Hilliard model. The scheme is based on the second-order backward differentiation formula in time and a finite difference method in space. Energy stability and convergence of the scheme are analyzed, and the optimal convergence orders in time and space are illustrated numerically. Note that the coefficient matrix of the scheme is a $2 \times 2$ block matrix with a Toeplitz-like structure in each block. Combining the advantages of this special structure with a Krylov subspace method, a preconditioning technique is designed to solve the system efficiently. Numerical examples are reported to illustrate the performance of the preconditioned iteration.

math.NA↗

arXiv2020

Generalized exceptional quantum walk search

We mainly study exceptional configuration for coined quantum walk search. For searching on a two-dimensional grid by AKR algorithm, we find some new classes of exceptional configurations that cannot be found by the AKR algorithm effectively and the known diagonal configuration can be regarded as its special case. Meanwhile, we give two modified quantum walk models that can improve the success probability in the exceptional configurations by numerical simulation. Furthermore, we introduce the concept of generalized exceptional configuration and consider search by quantum walk on a cycle with Grover coin. We find that the most natural coin combination model (G,-), where G is a Grover diffusion transformation, is a generalized exceptional configuration when just searching one marked vertex on the cycle. In the end, we find generalized exceptional configuration has a different evolution of quantum coherence from exceptional configuration. These extend largely the range of exceptional configuration of quantum walk search in some sense.

quant-ph↗

arXiv2020

A gridded establishment dataset as a proxy for economic activity in China

Measuring the geographical distribution of economic activity plays a key role in scientific research and policymaking. However, previous studies and data on economic activity either have a coarse spatial resolution or cover a limited time span, and the high-resolution characteristics of socioeconomic dynamics are largely unknown. Here, we construct a dataset on the economic activity of mainland China, the gridded establishment dataset (GED), which measures the volume of establishments at a 0.01$^{\circ}$ latitude by 0.01$^{\circ}$ longitude scale. Specifically, our dataset captures the geographically based opening and closing of approximately 25.5 million firms that registered in mainland China over the period 2005-2015. The characteristics of fine granularity and long-term observability give the GED a high application value. The dataset not only allows us to quantify the spatiotemporal patterns of the establishments, urban vibrancy and socioeconomic activity, but also helps us uncover the fundamental principles underlying the dynamics of industrial and economic development.

cs.SI↗

arXiv2020

Multi-Vehicle Routing Problems with Soft Time Windows: A Multi-Agent Reinforcement Learning Approach

Multi-vehicle routing problem with soft time windows (MVRPSTW) is an indispensable constituent in urban logistics distribution systems. Over the past decade, numerous methods for MVRPSTW have been proposed, but most are based on heuristic rules that require a large amount of computation time. With the current rapid increase of logistics demands, traditional methods incur the dilemma between computational efficiency and solution quality. To efficiently solve the problem, we propose a novel reinforcement learning algorithm called the Multi-Agent Attention Model that can solve routing problem instantly benefit from lengthy offline training. Specifically, the vehicle routing problem is regarded as a vehicle tour generation process, and an encoder-decoder framework with attention layers is proposed to generate tours of multiple vehicles iteratively. Furthermore, a multi-agent reinforcement learning method with an unsupervised auxiliary network is developed for the model training. By evaluated on four synthetic networks with different scales, the results demonstrate that the proposed method consistently outperforms Google OR-Tools and traditional methods with little computation time. In addition, we validate the robustness of the well-trained model by varying the number of customers and the capacities of vehicles.

cs.AI↗

arXiv2020

A PDE--based Power Tracking Control of Heterogeneous TCL Populations

This chapter presents the development and the analysis of a scheme for aggregate power tracking control of heterogeneous populations of thermostatically controlled loads (TCLs) based on partial differential equations (PDEs) control theory and techniques. By employing a thermostat--based deadband control with forced switching in the operation of individual TCLs, the aggregated dynamics of TCL populations are governed by a pair of Fokker--Planck equations coupled via the actions located both on the boundaries and in the domain. The technique of input-output feedback linearization is used for the design of aggregate power tracking control, which results in a nonlinear system in a closed loop. As the considered setting is a problem with time-varying boundaries, well-posedness assessment and stability analysis are carried out to confirm the validity of the developed control scheme. A simulation study is contacted to evaluate the effectiveness and the performance of the proposed approach.

math.OC↗

arXiv2020

Rhythmic Control of Automated Traffic -- Part II: Grid Network Rhythm and Online Routing

Connected and automated vehicle (CAV) technology is providing urban transportation managers tremendous opportunities for better operation of urban mobility systems. However, there are significant challenges in real-time implementation, as the computational time of the corresponding operations optimization model increases exponentially with increasing vehicle numbers. Following the companion paper (Chen et al., 2020) which proposes a novel automated traffic control scheme for isolated intersections, this study proposes a network-level real-time traffic control framework for CAVs on grid networks. The proposed framework integrates a rhythmic control (RC) method with an online routing algorithm to realize collisionfree control of all CAVs on a network and achieve superior performance in average vehicle delay, network traffic throughput, and computational scalability. Specifically, we construct a preset network rhythm that all CAVs can follow to move on the network and avoid collisions at all intersections. Based on the network rhythm, we then formulate online routing for the CAVs as a mixed integer linear program, which optimizes the entry times of CAVs at all entrances of the network and their time-space routings in real time. We provide a sufficient condition that the linear programming relaxation of the online routing model yields an optimal integer solution. Extensive numerical tests are conducted to show the performance of the proposed operations management framework under various scenarios. It is illustrated that the framework is capable of achieving negligible delays and increased network throughput. Furthermore, the computational time results are also promising. The CPU time for solving a collision-free control optimization problem with 2,000 vehicles is only 0.3 s on an ordinary personal computer.

math.OC↗

arXiv2020

Rhythmic Control of Automated Traffic -- Part I: Concept and Properties at Isolated Intersections

Leveraging the accuracy and consistency of vehicle motion control enabled by the connected and automated vehicle technology, we propose the rhythmic control (RC) scheme that allows vehicles to pass through an intersection in a conflict-free manner with a preset rhythm. The rhythm enables vehicles to proceed at a constant speed without any stop. The RC is capable of breaking the limitation that right of way can only be allocated to non-conflicting movements at a time. It significantly improves the performance of intersection control for automated traffic. Moreover, the RC with a predetermined rhythm does not require intensive computational efforts to dynamically control vehicles, which may possibly lead to frequent accelerations or decelerations. Assuming stationary vehicle arrivals, we conduct theoretical investigation to show that RC can considerably increase intersection capacity and reduce vehicle delay. Finally, the performance of RC is tested in the simulations with both stationary and non-stationary vehicle arrivals at both symmetric and asymmetric intersections.

math.OC↗

arXiv2020

NASGEM: Neural Architecture Search via Graph Embedding Method

Neural Architecture Search (NAS) automates and prospers the design of neural networks. Estimator-based NAS has been proposed recently to model the relationship between architectures and their performance to enable scalable and flexible search. However, existing estimator-based methods encode the architecture into a latent space without considering graph similarity. Ignoring graph similarity in node-based search space may induce a large inconsistency between similar graphs and their distance in the continuous encoding space, leading to inaccurate encoding representation and/or reduced representation capacity that can yield sub-optimal search results. To preserve graph correlation information in encoding, we propose NASGEM which stands for Neural Architecture Search via Graph Embedding Method. NASGEM is driven by a novel graph embedding method equipped with similarity measures to capture the graph topology information. By precisely estimating the graph distance and using an auxiliary Weisfeiler-Lehman kernel to guide the encoding, NASGEM can utilize additional structural information to get more accurate graph representation to improve the search efficiency. GEMNet, a set of networks discovered by NASGEM, consistently outperforms networks crafted by existing search methods in classification tasks, i.e., with 0.4%-3.6% higher accuracy while having 11%- 21% fewer Multiply-Accumulates. We further transfer GEMNet for COCO object detection. In both one-stage and twostage detectors, our GEMNet surpasses its manually-crafted and automatically-searched counterparts.

cs.AI↗

arXiv2020

Modeling light-controlled actuation of flexible magnetic composite structures using the finite element method (FEM)

Photoactive materials hold great promise for a variety of applications. We present a finite element model of light-controlled flexible magnetic composite structure composed of 33.3% Chromium dioxide (CrO2) and 66.7% Polydimethylsiloxane (PDMS) by weight. The structure has a dimension of 8 mm x 2 mm x 100 um and has been previously experimentally studied. Due to the low Curie temperature, the structure acts as an actuator, shows significant deflection under the external magnetic field and relaxation due to laser heating. Thermal and magnetic deflection analysis has been performed using the FEM model. The simulation results show a maximum structural deflection of 6.08 mm (76% of the length of the structure) when subjected to 30 mT magnetic flux density and 160 mW laser power at 303 K (room temperature). We will present the results of the simulation model and comparison to experimental data reproducing the previously observed motion of the (CrO2+PDMS). This model will enable future fracture and fatigue analysis as well as extension to new photoactive geometries.

physics.app-ph↗

arXiv2020

Simulation Comparisons of Vehicle-based and Phase-based Traffic Control for Autonomous Vehicles at Isolated Intersections

With the advent of autonomous driving technologies, traffic control at intersections is expected to experience revolutionary changes. Various novel intersection control methods have been proposed in the existing literature, and they can be roughly divided into two categories: vehicle-based traffic control and phase-based traffic control. Phase-based traffic control can be treated as updated versions of the current intersection signal control with the incorporation of the performance of autonomous vehicle functions. Meanwhile, vehicle-based traffic control utilizes some brand-new methods, mostly in real-time fashion, to organize traffic at intersections for safe and efficient vehicle passages. However, to date, no systematic comparison between these two control categories has been performed to suggest their advantages and disadvantages. This paper conducts a series of numerical simulations under various traffic scenarios to perform a fair comparison of their performances. Specifically, we allow trajectory adjustments of incoming vehicles under phasebased traffic control, while for its vehicle-based counterpart, we implement two strategies, i.e., the first-come-first-serve strategy and the conflict-point based rolling-horizon optimization strategy. Overall, the simulation results show that vehicle-based traffic control generally incurs a negligible delay when traffic demand is low but lead to an excessive queuing time as the traffic volume becomes high. However, performance of vehicle-based traffic control may benefit from reduction in conflicting vehicle pairs. We also discovered that when autonomous driving technologies are not mature, the advantages of phase-based traffic control are much more distinct.

eess.SY↗