arXiv subjects

Ke Ma

Publications and source records attributed to Ke Ma.

124 records · Page 7Linked to original sources

arXiv2021

On Stochastic Variance Reduced Gradient Method for Semidefinite Optimization

The low-rank stochastic semidefinite optimization has attracted rising attention due to its wide range of applications. The nonconvex reformulation based on the low-rank factorization, significantly improves the computational efficiency but brings some new challenge to the analysis. The stochastic variance reduced gradient (SVRG) method has been regarded as one of the most effective methods. SVRG in general consists of two loops, where a reference full gradient is first evaluated in the outer loop and then used to yield a variance reduced estimate of the current gradient in the inner loop. Two options have been suggested to yield the output of the inner loop, where Option I sets the output as its last iterate, and Option II yields the output via random sampling from all the iterates in the inner loop. However, there is a significant gap between the theory and practice of SVRG when adapted to the stochastic semidefinite programming (SDP). SVRG practically works better with Option I, while most of existing theoretical results focus on Option II. In this paper, we fill this gap via exploiting a new semi-stochastic variant of the original SVRG with Option I adapted to the semidefinite optimization. Equipped with this, we establish the global linear submanifold convergence (i.e., converging exponentially fast to a submanifold of a global minimum under the orthogonal group action) of the proposed SVRG method, given a provable initialization scheme and under certain smoothness and restricted strongly convex assumptions. Our analysis includes the effects of the mini-batch size and update frequency in the inner loop as well as two practical step size strategies, the fixed and stabilized Barzilai-Borwein step sizes. Some numerical results in matrix sensing demonstrate the efficiency of proposed SVRG method outperforming Option II counterpart as well as others.

math.OC↗

arXiv2020

Intrinsic Decomposition of Document Images In-the-Wild

Automatic document content processing is affected by artifacts caused by the shape of the paper, non-uniform and diverse color of lighting conditions. Fully-supervised methods on real data are impossible due to the large amount of data needed. Hence, the current state of the art deep learning models are trained on fully or partially synthetic images. However, document shadow or shading removal results still suffer because: (a) prior methods rely on uniformity of local color statistics, which limit their application on real-scenarios with complex document shapes and textures and; (b) synthetic or hybrid datasets with non-realistic, simulated lighting conditions are used to train the models. In this paper we tackle these problems with our two main contributions. First, a physically constrained learning-based method that directly estimates document reflectance based on intrinsic image formation which generalizes to challenging illumination conditions. Second, a new dataset that clearly improves previous synthetic ones, by adding a large range of realistic shading and diverse multi-illuminant conditions, uniquely customized to deal with documents in-the-wild. The proposed architecture works in a self-supervised manner where only the synthetic texture is used as a weak training signal (obviating the need for very costly ground truth with disentangled versions of shading and reflectance). The proposed approach leads to a significant generalization of document reflectance estimation in real scenes with challenging illumination. We extensively evaluate on the real benchmark datasets available for intrinsic image decomposition and document shadow removal tasks. Our reflectance estimation scheme, when used as a pre-processing step of an OCR pipeline, shows a 26% improvement of character error rate (CER), thus, proving the practical applicability.

cs.CV↗

arXiv2020

Attribute-guided image generation from layout

Recent approaches have achieved great success in image generation from structured inputs, e.g., semantic segmentation, scene graph or layout. Although these methods allow specification of objects and their locations at image-level, they lack the fidelity and semantic control to specify visual appearance of these objects at an instance-level. To address this limitation, we propose a new image generation method that enables instance-level attribute control. Specifically, the input to our attribute-guided generative model is a tuple that contains: (1) object bounding boxes, (2) object categories and (3) an (optional) set of attributes for each object. The output is a generated image where the requested objects are in the desired locations and have prescribed attributes. Several losses work collaboratively to encourage accurate, consistent and diverse image generation. Experiments on Visual Genome dataset demonstrate our model's capacity to control object-level attributes in generated images, and validate plausibility of disentangled object-attribute representation in the image generation from layout task. Also, the generated images from our model have higher resolution, object classification accuracy and consistency, as compared to the previous state-of-the-art.

cs.CV↗

arXiv2020

Incentive Compatibility in Stochastic Dynamic Systems

While the classic Vickrey-Clarke-Groves mechanism ensures incentive compatibility for a static one-shot game, it does not appear to be feasible to construct a dominant truth-telling mechanism for agents that are stochastic dynamic systems. The contribution of this paper is to show that for a set of LQG agents a mechanism consisting of a sequence of layered payments over time decouples the intertemporal effect of current bids on future payoffs and ensures truth-telling of dynamic states by their agents, if system parameters are known and agents are rational. Additionally, it is shown that there is a "Scaled" VCG mechanism that simultaneously satisfies incentive compatibility, social efficiency, budget balance as well as individual rationality under a certain "market power balance" condition where no agent is too negligible or too dominant. A further desirable property is that the SVCG payments converge to the Lagrange payments, the payments corresponding to the true price in the absence of strategic considerations, as the number of agents in the market increases. For LQ but non-Gaussian agents the optimal social welfare over the class of linear control laws is achieved.

eess.SY↗

arXiv2019

Fast Stochastic Ordinal Embedding with Variance Reduction and Adaptive Step Size

Learning representation from relative similarity comparisons, often called ordinal embedding, gains rising attention in recent years. Most of the existing methods are based on semi-definite programming (\textit{SDP}), which is generally time-consuming and degrades the scalability, especially confronting large-scale data. To overcome this challenge, we propose a stochastic algorithm called \textit{SVRG-SBB}, which has the following features: i) achieving good scalability via dropping positive semi-definite (\textit{PSD}) constraints as serving a fast algorithm, i.e., stochastic variance reduced gradient (\textit{SVRG}) method, and ii) adaptive learning via introducing a new, adaptive step size called the stabilized Barzilai-Borwein (\textit{SBB}) step size. Theoretically, under some natural assumptions, we show the $\boldsymbol{O}(\frac{1}{T})$ rate of convergence to a stationary point of the proposed algorithm, where $T$ is the number of total iterations. Under the further Polyak-Łojasiewicz assumption, we can show the global linear convergence (i.e., exponentially fast converging to a global optimum) of the proposed algorithm. Numerous simulations and real-world data experiments are conducted to show the effectiveness of the proposed algorithm by comparing with the state-of-the-art methods, notably, much lower computational cost with good prediction performance.

cs.LG↗

arXiv2019

The Strategic LQG System: A Dynamic Stochastic VCG Framework for Optimal Coordination

The classic Vickrey-Clarke-Groves (VCG) mechanism ensures incentive compatibility, i.e., that truth-telling of all agents is a dominant strategy, for a static one-shot game. However, in a dynamic environment that unfolds over time, the agents' intertemporal payoffs depend on the expected future controls and payments, and a direct extension of the VCG mechanism is not sufficient to guarantee incentive compatibility. In fact, it does not appear to be feasible to construct mechanisms that ensure the dominance of dynamic truth-telling for agents comprised of general stochastic dynamic systems. The contribution of this paper is to show that such a dynamic stochastic extension does exist for the special case of Linear-Quadratic-Gaussian (LQG) agents with a careful construction of a sequence of layered payments over time. For a set of LQG agents, we propose a modified layered version of the VCG mechanism for payments that decouples the intertemporal effect of current bids on future payoffs, and prove that truth-telling of dynamic states forms a dominant strategy if system parameters are known and agents are rational. An important example of a problem needing such optimal dynamic coordination of stochastic agents arises in power systems where an Independent System Operator (ISO) has to ensure balance of generation and consumption at all time instants, while ensuring social optimality. The challenge is to determine a bidding scheme between all agents and the ISO that maximizes social welfare, while taking into account the stochastic dynamic models of agents, since renewable energy resources such as solar/wind are stochastic and dynamic in nature, as are consumptions by loads which are influenced by factors such as local temperatures and thermal inertias of facilities.

eess.SY↗

arXiv2018

Less but Better: Generalization Enhancement of Ordinal Embedding via Distributional Margin

In the absence of prior knowledge, ordinal embedding methods obtain new representation for items in a low-dimensional Euclidean space via a set of quadruple-wise comparisons. These ordinal comparisons often come from human annotators, and sufficient comparisons induce the success of classical approaches. However, collecting a large number of labeled data is known as a hard task, and most of the existing work pay little attention to the generalization ability with insufficient samples. Meanwhile, recent progress in large margin theory discloses that rather than just maximizing the minimum margin, both the margin mean and variance, which characterize the margin distribution, are more crucial to the overall generalization performance. To address the issue of insufficient training samples, we propose a margin distribution learning paradigm for ordinal embedding, entitled Distributional Margin based Ordinal Embedding (\textit{DMOE}). Precisely, we first define the margin for ordinal embedding problem. Secondly, we formulate a concise objective function which avoids maximizing margin mean and minimizing margin variance directly but exhibits the similar effect. Moreover, an Augmented Lagrange Multiplier based algorithm is customized to seek the optimal solution of \textit{DMOE} effectively. Experimental studies on both simulated and real-world datasets are provided to show the effectiveness of the proposed algorithm.

cs.LG↗

arXiv2018

Robust Ordinal Embedding from Contaminated Relative Comparisons

Existing ordinal embedding methods usually follow a two-stage routine: outlier detection is first employed to pick out the inconsistent comparisons; then an embedding is learned from the clean data. However, learning in a multi-stage manner is well-known to suffer from sub-optimal solutions. In this paper, we propose a unified framework to jointly identify the contaminated comparisons and derive reliable embeddings. The merits of our method are three-fold: (1) By virtue of the proposed unified framework, the sub-optimality of traditional methods is largely alleviated; (2) The proposed method is aware of global inconsistency by minimizing a corresponding cost, while traditional methods only involve local inconsistency; (3) Instead of considering the nuclear norm heuristics, we adopt an exact solution for rank equality constraint. Our studies are supported by experiments with both simulated examples and real-world data. The proposed framework provides us a promising tool for robust ordinal embedding from the contaminated comparisons.

cs.LG↗

arXiv2018

Towards Provably Safe Mixed Transportation Systems with Human-driven and Automated Vehicles

Currently, we are in an environment where the fraction of automated vehicles is negligibly small. We anticipate that this fraction will increase in coming decades before if ever, we have a fully automated transportation system. Motivated by this we address the problem of provable safety of mixed traffic consisting of both intelligent vehicles (IVs) as well as human-driven vehicles (HVs). An important issue that arises is that such mixed systems may well have lesser throughput than all human traffic systems if the automated vehicles are expected to remain provably safe with respect to human traffic. This necessitates the consideration of strategies such as platooning of automated vehicles in order to increase the throughput. In this paper, we address the design of provably safe systems consisting of a mix of automated and human-driven vehicles including the use of platooning by automated vehicles. We design motion planing policies and coordination rules for participants in this novel mixed system. HVs are considered as nearsighted and modeled with relatively loose constraints, while IVs are considered as capable of following much tighter constraints. HVs are expected to follow reasonable and simple rules. IVs are designed to move under a model predictive control (MPC) based motion plans and coordination protocols. Our contribution of this paper is in showing how to integrate these two types of models safely into a mixed system. System safety is proved in single lane scenarios, as well as in multi-lane situations allowing lane changes.

eess.SY↗

arXiv2018

The Wide Integral Field Infrared Spectrograph: Commissioning Results and On-sky Performance

We have recently commissioned a novel infrared ($0.9-1.7$ $μ$m) integral field spectrograph (IFS) called the Wide Integral Field Infrared Spectrograph (WIFIS). WIFIS is a unique instrument that offers a very large field-of-view (50$^{\prime\prime}$ x 20$^{\prime\prime}$) on the 2.3-meter Bok telescope at Kitt Peak, USA for seeing-limited observations at moderate spectral resolving power. The measured spatial sampling scale is $\sim1\times1^{\prime\prime}$ and its spectral resolving power is $R\sim2,500$ and $3,000$ in the $zJ$ ($0.9-1.35$ $μ$m) and $H_{short}$ ($1.5-1.7$ $μ$m) modes, respectively. WIFIS's corresponding etendue is larger than existing near-infrared (NIR) IFSes, which are mostly designed to work with adaptive optics systems and therefore have very narrow fields. For this reason, this instrument is specifically suited for studying very extended objects in the near-infrared such as supernovae remnants, galactic star forming regions, and nearby galaxies, which are not easily accessible by other NIR IFSes. This enables scientific programs that were not originally possible, such as detailed surveys of a large number of nearby galaxies or a full accounting of nucleosynthetic yields of Milky Way supernova remnants. WIFIS is also designed to be easily adaptable to be used with larger telescopes. In this paper, we report on the overall performance characteristics of the instrument, which were measured during our commissioning runs in the second half of 2017. We present measurements of spectral resolving power, image quality, instrumental background, and overall efficiency and sensitivity of WIFIS and compare them with our design expectations. Finally, we present a few example observations that demonstrate WIFIS's full capability to carry out infrared imaging spectroscopy of extended objects, which is enabled by our custom data reduction pipeline.

astro-ph.IM↗

arXiv2018

Drag force in strongly coupled $\mathcal{N}=4$ SYM plasma in a magnetic field

Applying AdS/CFT correspondence, we study the effect of a constant magnetic field $\mathcal{B}$ on the drag force associated with a heavy quark moving through a strongly-coupled $\mathcal{N}=4$ supersymmetric Yang-Mills (SYM) plasma. The quark is considered moving transverse and parallel to $\mathcal{B}$, respectively. It is shown that for transverse case, the drag force is linearly dependent on $\mathcal{B}$ in all regions. While for parallel case, the drag force increases monotonously with increasing $\mathcal{B}$ and also reveals a linear behavior in the regions of strong $\mathcal{B}$. In addition, we find that $\mathcal{B}$ has important effect for transverse case than parallel.

hep-th↗

arXiv2018

Finding Global Optima in Nonconvex Stochastic Semidefinite Optimization with Variance Reduction

There is a recent surge of interest in nonconvex reformulations via low-rank factorization for stochastic convex semidefinite optimization problem in the purpose of efficiency and scalability. Compared with the original convex formulations, the nonconvex ones typically involve much fewer variables, allowing them to scale to scenarios with millions of variables. However, it opens a new challenge that under what conditions the nonconvex stochastic algorithms may find the global optima effectively despite their empirical success in applications. In this paper, we provide an answer that a stochastic gradient descent method with variance reduction, can be adapted to solve the nonconvex reformulation of the original convex problem, with a \textit{global linear convergence}, i.e., converging to a global optimum exponentially fast, at a proper initial choice in the restricted strongly convex case. Experimental studies on both simulation and real-world applications on ordinal embedding are provided to show the effectiveness of the proposed algorithms.

math.OC↗

arXiv2018

Stochastic Non-convex Ordinal Embedding with Stabilized Barzilai-Borwein Step Size

Learning representation from relative similarity comparisons, often called ordinal embedding, gains rising attention in recent years. Most of the existing methods are batch methods designed mainly based on the convex optimization, say, the projected gradient descent method. However, they are generally time-consuming due to that the singular value decomposition (SVD) is commonly adopted during the update, especially when the data size is very large. To overcome this challenge, we propose a stochastic algorithm called SVRG-SBB, which has the following features: (a) SVD-free via dropping convexity, with good scalability by the use of stochastic algorithm, i.e., stochastic variance reduced gradient (SVRG), and (b) adaptive step size choice via introducing a new stabilized Barzilai-Borwein (SBB) method as the original version for convex problems might fail for the considered stochastic \textit{non-convex} optimization problem. Moreover, we show that the proposed algorithm converges to a stationary point at a rate $\mathcal{O}(\frac{1}{T})$ in our setting, where $T$ is the number of total iterations. Numerous simulations and real-world data experiments are conducted to show the effectiveness of the proposed algorithm via comparing with the state-of-the-art methods, particularly, much lower computational cost with good prediction performance.

stat.ML↗

arXiv2015

A Theory for the Operation of the Independent System Operator in a Smart Grid with Stochastic Renewables, Demand Response and Storage

In this paper, we address a key issue of designing architectures and algorithms which generate optimal demand response in a decentralized manner for a smart-grid consisting of several stochastic renewables and dynamic loads. By optimal demand response, we refer to the demand response which maximizes the utility of the agents connected to the smart-grid. By decentralized we refer to the desirable case where neither the independent system operator (ISO) needs to know the dynamics/utilities of the agents, nor do the agents need to have a knowledge of the dynamics/utilities of other agents connected to the grid. The communication between the ISO and agents is restricted to the ISO announcing a pricing policy and the agents responding with their energy generation/consumption bids in response to the pricing policy. We provide a complete solution for both the deterministic and stochastic cases. It features a price iteration scheme that results in optimality of social welfare. We also provide an optimal solution for the case where there is a common randomness affecting and observed by all agents. This solution can be computationally complex, and we pose approximations. For the more general partially observed randomness case, we exhibit a relaxation that significantly reduces complexity. We also provide an approximation strategy that leads to a model predictive control (MPC) approach. Simulation results comparing the resulting optimal demand response with the existing architectures employed by the ISO illustrate the benefit in social welfare utility realized by our scheme. To the best of the authors' knowledge, this is the first work of its kind to explicitly mark out the optimal response of dynamic demand.

eess.SY↗

arXiv2014

A description of pseudorapidity distributions in p-p collisions at center-of-mass energy from 23.6 to 900 GeV

In the context of combined model of evolution-dominated hydrodynamics + leading particles, we discuss the pseudorapidity distributions of charged particles produced in p-p collisions. A comparison is made between the theoretical predictions and experimental measurements. The combined model works well in p-p collisions in the whole available energy region from sqrt(s_NN)=23.6 to 900 GeV.

nucl-th↗

arXiv2012

The Development of WIFIS: a Wide Integral Field Infrared Spectrograph

We present the current results from the development of a wide integral field infrared spectrograph (WIFIS). WIFIS offers an unprecedented combination of etendue and spectral resolving power for seeing-limited, integral field observations in the 0.9-1.8 um range and is most sensitive in the 0.9-1.35 um range. Its optical design consists of front-end re-imaging optics, an all-reflective image slicer-type, integral field unit (IFU) called FISICA, and a long-slit grating spectrograph back-end that is coupled with a HAWAII 2RG focal plane array. The full wavelength range is achieved by selecting between two different gratings. By virtue of its re-imaging optics, the spectrograph is quite versatile and can be used at multiple telescopes. The size of its field-of-view is unrivalled by other similar spectrographs, offering a 4.5" x 12" integral field at a 10-meter class telescope (or 20" x 50" at a 2.3-meter telescope). The use of WIFIS will be crucial in astronomical problems which require wide-field, two-dimensional spectroscopy such as the study of merging galaxies at moderate redshift and nearby star/planet-forming regions and supernova remnants. We discuss the final optical design of WIFIS, and its predicted on-sky performance on two reference telescope platforms: the 2.3-m Steward Bok telescope and the 10.4-m Gran Telescopio Canarias. We also present the results from our laboratory characterization of FISICA. IFU properties such as magnification, field-mapping, and slit width along the entire slit length were measured by our tests. The construction and testing of WIFIS is expected to be completed by early 2013. We plan to commission the instrument at the 2.3-m Steward Bok telescope at Kitt Peak, USA in Spring 2013.

astro-ph.IM↗