Mental workload (MWL) assessment is critical for accident prevention and operator safety. However, achieving cross-task generalization of MWL classification models is a significant challenge for real-world applications. Classifiers trained on labeled samples from one task often experience a notable performance drop when directly applied to samples from other tasks, limiting its use cases. To address this issue, we propose a semi-supervised cross-task domain adaptation (SCDA) method using power spectral density (PSD) features for MWL recognition across tasks (MATB-II and n-back). Our results demonstrated that the SCDA method achieved the best cross-task classification performance on our data and COG-BCI public dataset, with accuracies of 90.98% ± 9.36% and 96.61% ± 4.35%, respectively. Furthermore, in the cross-task classification of cross-subject scenarios, SCDA showed the highest average accuracy (75.39% ± 9.56% on our data, 90.98% ± 9.36% on the COG-BCI public dataset). The findings indicate that the semi-supervised transfer learning approach using PSD features is feasible and effective for cross-task MWL assessment.

Handwritten mathematical expression recognition (HMER) is an essential task in the OCR community, which consists of two sub-tasks, i.e., symbol recognition and structure parsing. Modern literature treats HMER as a LaTeX sequence predicting problem that simultaneously recognizes symbols and parses the structures of MEs. Although deep learning-based HMER methods have been achieving promising results on public benchmarks, it is admitted that the misclassification error between visually similar symbols still prevents these approaches from more generalized scenes. In this paper, we try to solve this issue from three aspects. 1) We enhanced the feature extraction progress by introducing path signature features, which incorporates local writing details and global spatial information. 2) We developed a language model that uses contextual information to correct the symbols misclassified by vision-only-based recognition models. 3) We solved the misalignment problem in existing ensemble method by designing a dynamic time warping (DTW) based algorithm. By combining the above improvements, our method achieved state-of-the-art results on three CROHME benchmarks, outperforming previous methods by a large margin.

Visual information extraction (VIE) has attracted increasing attention in recent years. The existing methods usually first organized optical character recognition (OCR) results in plain texts and then utilized token-level category annotations as supervision to train a sequence tagging model. However, it expends great annotation costs and may be exposed to label confusion, the OCR errors will also significantly affect the final performance. In this paper, we propose a unified weakly-supervised learning framework called TCPNet (Tag, Copy or Predict Network), which introduces 1) an efficient encoder to simultaneously model the semantic and layout information in 2D OCR results, 2) a weakly-supervised training method that utilizes only sequence-level supervision; and 3) a flexible and switchable decoder which contains two inference modes: one (Copy or Predict Mode) is to output key information sequences of different categories by copying a token from the input or predicting one in each time step, and the other (Tag Mode) is to directly tag the input sequence in a single forward pass. Our method shows new state-of-the-art performance on several public benchmarks, which fully proves its effectiveness.

For large-scale distributed systems, it is crucial to efficiently diagnose the root causes of incidents to maintain high system availability. The recent development of microservice architecture brings three major challenges (i.e., complexities of operation, system scale, and monitoring) to root cause analysis (RCA) in industrial settings. To tackle these challenges, in this paper, we present Groot, an event-graph-based approach for RCA. Groot constructs a real-time causality graph based on events that summarize various types of metrics, logs, and activities in the system under analysis. Moreover, to incorporate domain knowledge from site reliability engineering (SRE) engineers, Groot can be customized with user-defined events and domain-specific rules. Currently, Groot supports RCA among 5,000 real production services and is actively used by the SRE teams in eBay, a global e-commerce system serving more than 159 million active buyers per year. Over 15 months, we collect a data set containing labeled root causes of 952 real production incidents for evaluation. The evaluation results show that Groot is able to achieve 95% top-3 accuracy and 78% top-1 accuracy. To share our experience in deploying and adopting RCA in industrial settings, we conduct a survey to show that users of Groot find it helpful and easy to use. We also share the lessons learned from deploying and adopting Groot to solve RCA problems in production environments.

This paper studies the joint beamforming and power control in a multiuser multi-input single-output network by utilizing the only statistical channel distribution information. Such information consists of slowly varying covariance matrices in the beamforming network that can be employed to reduce instantaneous feedback overhead in transmission. Utilizing solely the statistical channel information, we study how to minimize the maximum outage probability under a weighted sum power constraint that guarantees max-min fairness to all users. This problem is, however, generally hard to solve due to the nonconvexity and nonlinear coupling between beamformer and power variables. First, assuming a fixed beamformer set, we use the nonlinear Perron-Frobenius theory to design a decentralized algorithm with provable geometrically fast convergence rate to compute the optimal power. Then, for the general case, we examine a certainty-equivalent margin counterpart with outage-mapped thresholds that incorporate the statistical channel information. We show that a network duality for this certainty-equivalent problem can be useful to decouple the coupling between the beamformer and power variables. This nonlinear Perron-Frobenius theory motivated approach yields a feasible beamformer and power allocation that are near-optimal as compared to Monte Carlo averaging simulations.

In this paper, we introduce a new approach to finger key point detection. For RGB images captured from an egocentric vision with a mobile camera, fingertip point detection remains a challenging problem due to various factors, like background complexity, illumination variety, hand shape diversity, and image blur cause by camera movements. To address these issues, we propose a bi-level cascade structure of a convolutional neuron network (CNN). The first-level CNN generates a bounding box of hand region by filtering a large proportion of complicated background information. Using the bounding box area as input, the second-level CNN including an extra branch returns accurate fingertip location with a multi-channel dataset. Our approach is the first attempt of finger key point detection from an egocentric vision with a mobile camera. The proposed method achieves satisfying and significant better results compared to previous fingertip detection methods based on handcraft features.

We consider the applications of blind and semi-blind interference alignment in multicell scenarios, specifically in clustered small cells. As a first step, two simple straight forward extensions of blind interference alignment are examined and it is observed that neither of them is uniformly superior. Then, we propose exploiting the location information of the users and base stations in the cluster to enhance the performance of fully blind schemes for any given user distribution scenario. Our aim is to group suitable users that can be served at the same time to minimize the supersymbol length for each cluster. Since the defined problem is NP-hard, we propose a heuristic algorithm that can provide an effective solution without too much complexity. By numerical simulations, we show that the proposed semi blind algorithm, Top.BIA, uniformly performs better than pure blind interference alignment schemes for any possible user distribution scenario.

In modern heterogeneous wireless networks, the task of supporting fairness along with user priorities and concurrently achieving the highest possible system throughput is desirable and challenging. Herein, a class of practical cumulative distribution function (CDF) scheduling algorithms are developed to achieve these goals. These algorithms are used when the channel fading model is unknown. The mapping from channel quality information (CQI) to the real CDF is unknown but is constructed exploiting the order statistics of the CQI sequence. The constructed CDF mapping methods are shown to converge to the actual CDF. Specifically, one algorithm uses the expected value of the ordered CDF scheduling while others called Non-parametric CDF scheduling (NPCS) algorithms reconstructs the CDF with an extra interpolation step. By collecting a moderate number of CQI data, the algorithms almost achieve the system throughput of CDF scheduling as if the CDF is known. Throughout the work, CDF scheduling algorithms, supported by simulations, are shown to be able to effectively support fairness and frequently outperform, and are potential alternatives to, the well known Proportional Fair (PF) scheduling method.

This paper studies the joint beamforming and power control in a multiuser distributed antenna uplink network, wherein the number of users and the number of separately located antennas grow large with ratio being bounded. We consider the SINR fairness problem under individual power constraint and present a distributed iterative algorithm. This algorithm, though converging to the instantaneous optimal solution, requires instantaneous power update. In order to design a low complexity algorithm that achieves optimality in the asymptotic sense, we leverage the large system structure to derive an asymptotic solution requiring only channel statistics. In this algorithm, the asymptotic power is slowly updated and the asymptotic beamformer can be obtained in a non-iterative manner. The convergence property of the proposed algorithm is also studied.

In a multicell multiuser MIMO downlink employing random beamforming as the transmission scheme, the heterogeneous large scale channel effects of intercell and intracell interference complicate analysis of distributed scheduling based systems. In this paper, we extend the analysis in [1] and [2] to study the aforementioned challenging scenario. The cumulative distribution function (CDF)-based scheduling policy utilized in [1] and [2] is leveraged to maintain fairness among users and simultaneously obtain multiuser diversity gain. The closed form expression of the individual sum rate for each user is derived under the CDF-based scheduling policy. More importantly, with this distributed scheduling policy, we conduct asymptotic (in users) analysis to determine the limiting distribution of the signal-to-interference-plus-noise ratio, and establish the individual scaling laws for each user.

In this paper, we show that optimizing the feedback strategy in a limited feedback multiuser system can significantly improve system performance. Herein, we consider a limited feedback multiuser system, in which the fairness among users are guaranteed by the application of the cumulative distribution function (CDF) based scheduling technique. As the users are diverse in types, priorities and channel statistics, adjusting feedback strategy as well as dynamically distributing feedback resources among the users helps the system to exploit multiuser diversity gain. The system performance optimization is formulated in which we consider feedback as a resource to optimize. Then, the optimization is decoupled into two sub-problems, namely optimization of quantizers and allocation of feedback bits for the users. Optimizing feedback usage is shown to be significantly beneficial, both analytically and with the help of simulation results. Finally, in order to address the practical aspect, we propose an algorithm to efficiently solve these two sub-problems in an optimal manner followed by a low complexity sub-optimal approach which offers good tradeoff between complexity and performance.

We consider a multiuser downlink system where the base station (BS) only has imperfect channel quality information (CQI) due to both feedback delay and estimation error at the receiver. Using coherent detection, the signal portion due to delay is accommodated while the estimation error is random and can be considered as noise. A thresholding feedback scheme is used to reduce the overall amount of feedback which is typically set by the system. Then, a fixed rate and a variable rate transmission scheme are analyzed which are then optimized to obtain the highest average system goodput. By using variable rate with optimized parameters, system goodput increases significantly in comparison with the fixed rate strategy. The feedback delay is shown to affect system performance with the severity increasing over time. Finally, the match between analysis and simulation are verified by the experiments.

This paper investigates three open problems in random beamforming based communication systems: the scheduling policy with heterogeneous users, the closed form sum rate, and the randomness of multiuser diversity with selective feedback. By employing the cumulative distribution function based scheduling policy, we guarantee fairness among users as well as obtain multiuser diversity gain in the heterogeneous scenario. Under this scheduling framework, the individual sum rate, namely the average rate for a given user multiplied by the number of users, is of interest and analyzed under different feedback schemes. Firstly, under the full feedback scheme, we derive the closed form individual sum rate by employing a decomposition of the probability density function of the selected user's signal-to-interference-plus-noise ratio. This technique is employed to further obtain a closed form rate approximation with selective feedback in the spatial dimension. The analysis is also extended to random beamforming in a wideband OFDMA system with additional selective feedback in the spectral dimension wherein only the best beams for the best-L resource blocks are fed back. We utilize extreme value theory to examine the randomness of multiuser diversity incurred by selective feedback. Finally, by leveraging the tail equivalence method, the multiplicative effect of selective feedback and random observations is observed to establish the individual rate scaling.

This paper studies joint beamforming and power control in a coordinated multicell downlink system that serves multiple users per cell to maximize the minimum weighted signal-to-interference-plus-noise ratio. The optimal solution and distributed algorithm with geometrically fast convergence rate are derived by employing the nonlinear Perron-Frobenius theory and the multicell network duality. The iterative algorithm, though operating in a distributed manner, still requires instantaneous power update within the coordinated cluster through the backhaul. The backhaul information exchange and message passing may become prohibitive with increasing number of transmit antennas and increasing number of users. In order to derive asymptotically optimal solution, random matrix theory is leveraged to design a distributed algorithm that only requires statistical information. The advantage of our approach is that there is no instantaneous power update through backhaul. Moreover, by using nonlinear Perron-Frobenius theory and random matrix theory, an effective primal network and an effective dual network are proposed to characterize and interpret the asymptotic solution.

This paper studies joint beamforming and power control in a multiuser MISO interference network with statistical channel information. Such information consists of the slow-varying covariance matrices in the beamforming network, and can be employed to reduce instantaneous feedback needs. With the outage event induced by the utilization of statistical channel information, we optimize signal transmission strategies to minimize the maximum outage probability under weighted sum power constraint to achieve outage balancing in the interference network. Under the condition of fixed beamformer, we use nonlinear Perron-Frobenius theory to present a decentralized algorithm with provable geometrically fast convergence rate to compute the optimal power. Since the joint beamformer and power optimization problem is non-convex, we examine its certainty-equivalent margin counterpart. By leveraging nonlinear Perron-Frobenius theory and the established network duality, we present a near-optimal decentralized algorithm to jointly optimize the beamformer and power. The algorithm converges quickly and the convergence rate of the algorithm is proven to be geometrical.

This paper studies efficient signal transmission strategies in a multiuser MIMO network where multiple source nodes form a large virtual MIMO array to perform cooperative beamforming. In order to reduce instantaneous feedback needs, we investigate joint power control and cooperative beamforming techniques to maximize the minimum weighted average SINR based on statistical channel information, which consists of the covariance matrices in the beamforming network. Since the joint optimization problem over power, transmit beamformer and receive beamformer is non-convex, we analyze two sub-problems: the joint optimization of power and transmit beamformer under fixed receive beamformer, and the joint optimization of power and receive beamformer under fixed transmit beamformer. For both sub-problems, we use nonlinear Perron-Frobenius theory to characterize the optimal solution and present decentralized algorithms with provable geometrically fast convergence rate. By combining the developed techniques and the network duality, we provide an effective iterative algorithm for the joint optimization problem.

Spatial and multiuser diversity are two types of diversity techniques for delivering reliable high-date-rate services. Spectral diversity comes from opportunistic scheduling in the frequency domain enabled by the OFDMA technique, and is influenced by partial feedback design. By employing the best-M partial feedback strategy, we provide a unified view of spatial, spectral, and multiuser diversity through asymptotic (in users) analysis. We examine the tail behavior of the distribution of the received channel quality information (CQI) at the scheduler to prove the type of convergence as well as to derive the asymptotic approximations for the average spectral efficiency under partial feedback. We investigate the application of our analysis to different spatial diversity schemes. Our derived results can be used to quickly determine the minimum required partial feedback in a general multiuser MIMO-OFDMA system.

Modern heterogeneous networks have an inherent heterogeneous structure such as user densities and large scale channel effects that motivates this work on heterogeneous partial feedback design. In this paper, we address the partial feedback design issue in an OFDMA-based macro-pico cellular system employing the best-M partial feedback strategy. We consider the scenario with one picocell inside a macrocell, and investigate a scheduling policy which tracks the small scale channel effects in order to guarantee fairness among users as well as exploiting the multiuser diversity. We derive a closed form expression for the average spectral efficiency of the network with best-M partial feedback for the identified representative scenarios. We carry out asymptotic analysis using extreme value theory to derive an approximation to the average spectral efficiency in order to determine the minimum required partial feedback with minimal loss in system performance.

We present a unified analytical framework to jointly examine spatial, spectral, and multiuser diversity in a general multiuser MIMO OFDMA system. Spectral diversity originates from frequency domain opportunistic scheduling enabled by the OFDMA technique, and is influenced by partial feedback design. We propose a heterogeneous partial feedback design method which adapts users' feedback resources to their frequency domain channel statistics. We investigate this adaptive feedback design by employing the best-M partial feedback strategy, and derive closed form expressions for the average spectral efficiency and bit error rate with different spatial diversity schemes under the generalized Nakagami-m fading channels. We also examine the interplay among the diversity effects using our heterogeneous feedback design as well as the channel fading effect through numerical results.

In this letter, we investigate the sum rate performance of a generic one-pico-inside OFDMA network with opportunistic scheduling policy and the best-M partial feedback design. By a detailed examination of the statistical property of the selected user's signal-to-interference-plus-noise ratio and its interplay with intercell interference, we derive the exact closed form expression for the sum rate under partial feedback. The general expression incorporates the special interference limited and noise limited scenarios as special cases. The result explains the exact interaction with partial feedback and the approach can be generalized and applied to many related problems.

Current communication schemes rely on instantaneous adaptation, which increases the system complexity and is memoryless. We propose a scheme for adapting system parameters based on a priori knowledge of the channel statistics to reduce feedback requirements. As a vital parameter in OFDM system, cyclic prefix (CP) length depends mainly on two channel characteristics: the channel gains and the root mean square (RMS) delay spread. Previous research assume given RMS delay spread and utilize power delay profile (PDP) for CP optimization. Since PDP indeed acts as a priori statistics of channel gains, we compare its effectiveness against a feedback scheme based on the instantaneous channel. We further simplify CP optimization by employing the prior statistics of the RMS delay spread. We derive a closed form approximation scheme for computation simplicity assuming RMS delay spread follows a lognormal distribution. We consider ergodic capacity and measure the capacity loss w.r.t. an instantaneous scheme. The observed 1% - 4% capacity loss through simulation validates our proposed scheme. We also present an online method to learn the prior statistics of RMS delay spread from feedback whenever available.