Facebook Marketplace is quickly gaining momentum among consumers as a favored customer-to-customer (C2C) product trading platform. The recommendation system behind it helps to significantly improve the user experience. Building the recommendation system for Facebook Marketplace is challenging for two reasons: 1) Scalability: the number of products in Facebook Marketplace is huge. Tens of thousands of products need to be scored and recommended within a couple hundred milliseconds for millions of users every day; 2) Cold start: the life span of the C2C products is very short and the user activities on the products are sparse. Thus it is difficult to accumulate enough product level signals for recommendation and we are facing a significant cold start issue. In this paper, we propose to address both the scalability and the cold-start issue by building a collaborative multi-modal deep learning based retrieval system where the compact embeddings for the users and the products are trained with the multi-modal content information. This system shows significant improvement over the benchmark in online and off-line experiments: In the online experiment, it increases the number of messages initiated by the buyer to the seller by +26.95%; in the off-line experiment, it improves the prediction accuracy by +9.58%.

A novel type of trivalent BNg five-membered cational species B5Ngn3+(Ng = He~Rn, n = 1~5) has been found and investigated theoretically using the B3LYP and MP2 methods with the def2-QZVPPD and def2-TZVPPD basis sets. The geometry, harmonic vibrational frequencies, bond energies, charge distribution, bond nature, aromaticity, and energy decomposition analysis of these structures were reported. The calculated B−Ng bond energy is quite large (the averaged bond energy is in the range of 209.2~585.76 kJ mol-1) for heavy rare gases and increases with the Ng atomic number. The analyses of the molecular wavefunction show that in the BNg compounds of heavy Ng atoms Ar~Rn, the B−Ng bonds are of typical covalent character. Nuclear independent chemical shifts display that both B53+ and B5Ngn3+(n=1~5) have obvious aromaticity. Energy decomposition analysis shows that these BNg compounds are mainly stabilized by the σ-donation from the Ng valence p orbital to the B53+ LUMO. These findings offer valuable clues toward the design and synthesis of new stable Ng-containing compounds.

Previous work on dialog act (DA) classification has investigated different methods, such as hidden Markov models, maximum entropy, conditional random fields, graphical models, and support vector machines. A few recent studies explored using deep learning neural networks for DA classification, however, it is not clear yet what is the best method for using dialog context or DA sequential information, and how much gain it brings. This paper proposes several ways of using context information for DA classification, all in the deep learning framework. The baseline system classifies each utterance using the convolutional neural networks (CNN). Our proposed methods include using hierarchical models (recurrent neural networks (RNN) or CNN) for DA sequence tagging where the bottom layer takes the sentence CNN representation as input, concatenating predictions from the previous utterances with the CNN vector for classification, and performing sequence decoding based on the predictions from the sentence CNN model. We conduct thorough experiments and comparisons on the Switchboard corpus, demonstrate that incorporating context information significantly improves DA classification, and show that we achieve new state-of-the-art performance for this task.

Phasor measurement units (PMUs) are becoming an important component in the power grid. The PMUs provide time-synchronized measurements of multiple remote measurement points, enabling more accurate and real-time monitoring of the system state. Heuristically, it is difficult to tell how the placement of a PMU at a specific location in the power system would affect the state estimation error, not to say the effect of the PMU synchronization and measurement accuracy. In this paper, we derive the posterior Cramer-Rao bound (PCRB) on the state estimation error based on a measurement model which considers the phase-angle mismatch from PMU measurements. We then propose a PMU placement strategy using the derived PCRB. The greedy algorithm is used to solve the optimization problem. The results are then compared with other heuristics, with the optimal solution through an exhaustive search (for small systems), and with a lower bound on the optimal placement obtained through convex relaxation. For some design criteria, the objective functions are submodular, which guarantees a performance bound on the greedy solution. For other design criteria where the objective functions are not submodular, numerical examples demonstrate the effectiveness of the greedy algorithms.

We consider the problem of direction of arrival (DOA) estimation using a newly proposed structure of co-prime arrays. A continuous sparse recovery method is implemented in order to increase resolution. We show that in the noiseless case one can theoretically detect up to MN/2 sources with only 2M+N sensors via continuous sparse recovery. The noise statistics of co-prime arrays are also analyzed to demonstrate the robustness of the proposed optimization scheme. Using numerical examples, we show the superiority of the proposed method.

In this letter, we consider the problem of direction of arrival estimation using sparsity enforced reconstruction methods. Co-prime arrays with $M + N$ sensors are utilized to increase the degrees of the freedom from $O(M + N)$ to $O(MN)$ . The key to the success of sparse-based direction of arrival estimation is that every target must fall on the predefined grid. Off-grid targets can highly jeopardize the reconstruction performance. In this letter, we use joint sparsity reconstruction methods to explore the underlying structure between the sparse signal and the gird mismatch. Two types of sparse reconstruction methods, the greedy method and the convex relaxation method, are considered. By implementing numerical experiments, we demonstrate that our proposed methods can fully utilize the virtual aperture created by co-prime arrays and also outperform the previously proposed MUSIC method with spatial smoothing.

The application of ultrasonic guided waves has been got considerable attentions in non-destructive evaluation (NDE). However, the mode overlap and dispersion problem of guided waves are still challenging. In the present study, we used the spectral phase and time delay function to model the guided dispersion transfer system, and developed a compensation algorithm of single mode dispersion. Aiming to overcome the unsolved difficulty of multimode processing, dispersion compensation was then combined with the mode separation. Simulations of the Lamb modes A0, A1 and S0 illustrated that the proposed algorithm is capable of synthesizing and compensating of single mode dispersion. Furthermore, using the selectively compensation, the individual modes can also be extracted from the multimodal guided signals. The study is helpful for the signal processing of multimodal guided dispersion.

In this chapter, we propose an optimal and distributed control strategy for plug-in electric vehicles’ (PEVs) charging as part of demand response in the smart grid. We consider an electricity market where users have the flexibility to sell back the energy stored in their PEVs or the energy generated from their distributed generators. The smart grid model in this chapter integrates a two-way communication system between the utility company and consumers. A price scheme considering fluctuation cost is developed to encourage consumers to lower the fluctuation in the demand response by charging and discharging their PEVs reasonably. A distributed optimization algorithm based on the alternating direction method of multipliers is applied to solve the optimization problem, in which consumers need to report their aggregated loads only to the utility company, thus ensuring their privacy. Consumers update the scheduling of their loads simultaneously and locally to speed up the optimization computing. We also extend the distributed algorithm to the asynchronous case, where communication loss exists in the smart grid. Using numerical examples, we show that the demand curve is flattened after the optimal PEV charging and load scheduling. We also show the robustness of the proposed method by considering estimation uncertainty on the overall next day load, and also the renewable energy. The distributed algorithms are shown to reduce the users’ daily bills with respect to different scenarios, thus motivating consumers to participate in the proposed framework.

In this paper, we propose a new model of demand response management for the future smart grid that integrates plug-in electric vehicles and renewable distributed generators. A price scheme considering fluctuation cost is developed. We consider a market where users have the flexibility to sell back the energy generated from their distributed generators or the energy stored in their plug-in electric vehicles. A distributed optimization algorithm based on the alternating direction method of multipliers is developed to solve the optimization problem, in which consumers need to report their aggregated loads only to the utility company, thus ensuring their privacy. Consumers can update their loads scheduling simultaneously and locally to speed up the optimization computing. Using numerical examples, we show that the demand curve is flattened after the optimization, even though there are uncertainties in the model, thus reducing the cost paid by the utility company. The distributed algorithms are also shown to reduce the users' daily bills.

In traditional compressed sensing theory, the dictionary matrix is given a priori, whereas in real applications this matrix suffers from random noise and fluctuations. In this paper, we consider a signal model where each column in the dictionary matrix is affected by a structured noise. This formulation is common in direction-of-arrival (DOA) estimation of off-grid targets, encountered in both radar systems and array processing. We propose to use joint sparse signal recovery to solve the compressed sensing problem with structured dictionary mismatches and also give an analytical performance bound on this joint sparse recovery. We show that, under mild conditions, the reconstruction error of the original sparse signal is bounded by both the sparsity and the noise level in the measurement model. Moreover, we implement fast first-order algorithms to speed up the computing process. Numerical examples demonstrate the good performance of the proposed algorithm and also show that the joint-sparse recovery method yields a better reconstruction result than existing methods. By implementing the joint sparse recovery method, the accuracy and efficiency of DOA estimation are improved in both passive and active sensing cases.

We consider the problem of direction of arrival (DOA) estimation using a newly proposed structure of non-uniform linear arrays, referred to as co-prime arrays, in this paper. By exploiting the second order statistical information of the received signals, co-prime arrays exhibit O(MN) degrees of freedom with only M + N sensors. A sparsity based recovery method is proposed to fully utilize these degrees of freedom. Unlike traditional sparse recovery methods, the proposed method is based on the developing theory of super resolution, which considers a continuous range of possible sources instead of discretizing this range into a discrete grid. With this approach, off-grid effects inherited in traditional sparse recovery can be neglected, thus improving the accuracy of DOA estimation. In this paper we show that in the noiseless case one can theoretically detect up to M N sources with only 2M + N sensors. The noise 2 statistics of co-prime arrays are also analyzed to demonstrate the robustness of the proposed optimization scheme. A source number detection method is presented based on the spectrum reconstructed from the sparse method. By extensive numerical examples, we show the superiority of the proposed method in terms of DOA estimation accuracy, degrees of freedom, and resolution ability compared with previous methods, such as MUSIC with spatial smoothing and the discrete sparse recovery method.

We consider algorithms and recovery guarantees for the analysis sparse model in which the signal is sparse with respect to a highly coherent frame. We consider the use of a monotone version of the fast iterative shrinkage-thresholding algorithm (MFISTA) to solve the analysis sparse recovery problem. Since the proximal operator in MFISTA does not have a closed-form solution for the analysis model, it cannot be applied directly. Instead, we examine two alternatives based on smoothing and decomposition transformations that relax the original sparse recovery problem, and then implement MFISTA on the relaxed formulation. We refer to these two methods as smoothing-based and decomposition-based MFISTA. We analyze the convergence of both algorithms and establish that smoothing-based MFISTA converges more rapidly when applied to general nonsmooth optimization problems. We then derive a performance bound on the reconstruction error using these techniques. The bound proves that our methods can recover a signal sparse in a redundant tight frame when the measurement matrix satisfies a properly adapted restricted isometry property. Numerical examples demonstrate the performance of our methods and show that smoothing-based MFISTA converges faster than the decomposition-based alternative in real applications, such as MRI image reconstruction.

Phasor measurement units (PMUs) are time synchronized sensors for power system state estimation. Despite their promising advantages, large scale deployment of PMUs is still limited. One of the reasons is the high cost of their exact synchronization. In this paper, we consider the use of cheaper and less accurately synchronized PMUs, but compensate for these imperfections via signal processing methods. We introduce a statistical model for power state estimation using these impaired units. We then derive an alternating minimization technique and a parallel Kalman filter for static and dynamic estimation, respectively. Numerical examples demonstrate the improvement in estimation accuracy of our algorithm compared with traditional algorithms when phase mismatch is present. Our results suggest that the phase mismatches can be largely compensated as long as there is a sufficient number of PMUs and the delays are small.

To drive the application of surface‐enhanced Raman spectroscopy (SERS) mapping in ex vivo diagnostic imaging and non‐biological material characterization, we have designed a robust and accurate multiplex spectral fitting method using an alternating minimization algorithm to extract individual constituent Raman spectra with very small overall fitting error (as low as 2%). For each mixed Raman signal, constituent spectra and mixture coefficients were estimated jointly based on reference spectra that were measured in the lab. Our method is based on a Poisson model to reflect the photon counting nature of Raman signals and includes the nonlinear noise in the measured data, making our method robust against data containing relatively large random noise. In our method, we minimized a cost function consisting of two terms: (1) the overall fitting error between the measured and modeled spectra and (2) the sum of the individual error between each reference spectrum and its corresponding constituent. This method inherently guarantees that the estimates will approach the global minimum with monotonic convergence. The accuracy of our method was validated by applying it to a SERS spectral fitting problem and comparing our results to those from existing methods. Copyright © 2013 John Wiley & Sons, Ltd.

Phasor measurement units (PMUs) are time synchronized sensors primarily used for power system state estimation. Despite their increasing incorporation and the ongoing research on state estimation using measurements from these sensors, estimation with imperfect phase synchronization has not been sufficiently investigated. Inaccurate synchronization is an inevitable problem that large scale deployment of PMUs has to face. In this paper, we introduce a model for power system state estimation using PMUs with phase mismatch. We propose alternating minimization and parallel Kalman filtering for state estimation using static and dynamic models, respectively, under different assumptions. Numerical examples demonstrate the improved accuracy of our algorithms compared with traditional algorithms when imperfect synchronization is present. We conclude that when a sufficient number of PMUs with small delays are employed, the imperfect synchronization can be largely compensated in the estimation stage.

In this paper we propose a sparse model to accurately estimate target locations in a distributed multiple-input multiple-output (MIMO) radar system with phase mismatches at transmitters and receivers. We formulate the localization problem based on maximum a posteriori (MAP) estimation. To reduce the effect of phase mismatches we develop a novel alternating minimization approach based on sparse signal recovery and structured matrix perturbation. Using numerical simulations, we show that our algorithms significantly improve the performance of the distributed MIMO radar system.

Phasor measurement units (PMU) provide time-synchronized linear measurements for power system state estimation. In practice, the synchronization is not perfect for various reasons. In this paper we derive the posterior Cramér-Rao bound on the estimation error based on a realistic measurement model which takes into account the synchronization error. We then use a greedy algorithm for PMU placement based on the derived bound, and compare the results with other heuristics and the optimal solution through exhaustive search. Numerical examples demonstrate the performance improvement using the PMU placement profile from the greedy algorithm. The results also indicate that the greedy technique closely approximates the optimal solution.

In traditional compressed sensing theory, the dictionary matrix is given a priori, while in real applications this matrix suffers from random noise and fluctuations. This paper considers a signal model where each column in the dictionary matrix is affected by a structured noise. This formulation is common in radar related applications and direction-of-arrival estimation. We propose to use joint-sparse signal recovery in this compressed sensing problem with dictionary mismatch and also give a theoretical result on the performance bound for this joint-sparse method. We show that under mild conditions the reconstruction error of the original sparse signal is bounded by both the sparsity and the noise level in the measurement model. Moreover, a fast first-order method is implemented to speed up the computing process. Numerical examples demonstrate the good performance of the proposed algorithm, and also show that the joint-sparse recovery method converges faster and gives a better reconstruction result than previous methods.

In this paper, we propose a new model of demand response management for the future smart grid that integrates plug-in electric vehicles. A price scheme considering fluctuation cost is developed. We consider a market where users have the flexibility to sell back the energy generated from their distributed generators or the energy stored in their plug-in electric vehicles. A distributed optimization algorithm based on the alternating direction method of multipliers is developed to solve the optimization problem, in which consumers need to report their aggregated load only to the utility company, thus ensuring their privacy. Consumers can update their load scheduling simultaneously and locally to speed up the optimization computing. Using numerical examples, we show the demand curve is flattened after the optimization, thus reducing the cost paid by the utility company. The distributed algorithm is also shown to reduce the users' daily bills.