Baixuan Li, Jialong Wu, Wenbiao Yin, Kuan Li, Zhongwang Zhang, Huifeng Yin, Zhengwei Tao, Liwen Zhang, Pengjun Xie, Jingren Zhou, Yong Jiang, Wentao Zhang, Zhiqiang Gao. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Modern compilers like LLVM and GCC provide hundreds of optimization options (e.g., flags or passes), yet their fixed, predefined sequences (e.g., -O3) often fail to exploit the full performance potential of specific programs. Search-based auto-tuning has emerged to address this pass selection and ordering problem — known as the phase ordering problem. Existing approaches typically reduce search complexity by identifying critical flags or employing localized group-based mutations. However, these methods often remain feature-agnostic or rely on manually designed static mappings during the online search. Consequently, they fail to adaptively link program features to optimization logic. Furthermore, they are prone to being misled by spurious correlations and noise inherent in sparse performance data.

To perform segmentation for the cell images of different modalites accurately, we should address issues of over-segmentation or under-segmentation caused by uncertainty variations in modal pixel distribution and cell morphology. Moreover, the problem of limited labeled data is studied in this work. Most previous methods lack global uncertainty information perception ability, can not obtain local uncertainty details, and are limited to the number of labeled data from multi-modal cell images. We introduce a novel framework that can accurately learn valuable information for multi-modal cell segmentation task with the data and modal uncertainty aware abilities. Firstly, an image fusion module is proposed that leverages a multi-branch structure, incorporating dilation convolution, regular convolution, and channel attention mechanism for saving global valuable information. Secondly, to obtain local boundaries from obscure and irregular uncertainty regions, a transformer-based model encoding strategy is developed for performing token selection and enhancement based on the feedback confidence score. This confidence score is computed based on the output of a teaching network that indicates most likely local boundary. Thirdly, a pseudo-label selection strategy is employed to improve the annotation quality of unlabeled images. We evaluated our method on three publicly available datasets with different cell modalites and performed a quantitative comparison with the previous fifteen methods. Our method achieved better performance than others. This study has important implications for the improvements of clinical applications including diagnostic accuracy and decision-making reliability.

Compiler developers have designed and implemented numerous compiler optimization techniques and provided several standard compilation flag combinations to simplify the user workflow. However, achieving the desired optimization results is difficult when using the standard compiler flag combinations provided or manually selecting the compiler flag combinations. Thus, we propose PreTuner, a general compilation flag optimization framework based on iterative compilation and predictive models. Using the optimization results from evolutionary algorithms to build a dataset of code features and compilation flag combinations, we can optimize any program in one round by extracting code features and combining them with compilation flags. PreTuner effectively addresses the problem of limited dataset availability and improves model generalization. Based on the GCC compiler, PreTuner achieves an average speedup of 1.098× (up to 1.252×) compared to the -O3 standard optimization flag combination on the cBench test suite, validating the feasibility and effectiveness of PreTuner.

The scheduling of aircraft support tasks requires efficient planning based on the available resources in each sup-port position. The many-to-many characteristics of tasks and the dynamic nature of scheduling environments place high demands on the real-time responsiveness of algorithms. Additionally, the complexity inherent in task scheduling and the need for flexible sequential processing further complicate decision-making. Existing methods often struggle to achieve both fast response times and effective task message capture. To address these challenges, we propose a novel scheduling method called Multi-DAT, which is designed to optimize dynamic task allocation using a multi-agent reinforcement learning algorithm. We improve on the traditional QMIX algorithm to select the shortest duration tasks based on priority, combined with a newly designed long-term reward function, which integrates long-term historical actions into the scheduling algorithm. Experimental results demonstrate that our proposed method outperforms the traditional rule-based method and seven other multi-agent reinforcement learning-based scheduling algorithms in terms of scheduling performance.

Eye movement classification can decode cognitive processes, offering valuable insights for a wide range of applications. However, existing eye movement classification models primarily focus on static fixation-based features and often neglect the encoding of spatiotemporal eye movement features, which are crucial for accurately reconstructing visual attention. To address this limitation, we propose a spatiotemporal attention encoding (STAE) model that jointly captures both spatial and temporal features for eye-tracking classification. First, we utilize a Vision Transformer (ViT) to extract spatial features from fixations by taking global competition into consideration. We then introduce a global weighting Gated Recurrent Unit (GRU) model to capture temporal correlations from the feature sequence. Specifically, we propose a hidden-state-based weighting to fuse the influence of different fixations on the current fixation. In the experiment, we evaluated our model on three tasks: autism spectrum disorder (ASD) identification, visual task classification, and age classification. Experiential results across three databases demonstrate that our model outperforms existing methods and shows strong adaptability across various eye movement classification tasks. The code is available at https://github.com/HectorTo/spatiotemporal-attention-encoding-STAE-.

As Large Language Model (LLM) context windows expand, the necessity of Retrieval-Augmented Generation (RAG) for integrating external knowledge is debated. Existing RAG vs. long-context (LC) LLM comparisons are often inconclusive due to benchmark limitations. We introduce LaRA, a novel benchmark with 2326 test cases across four QA tasks and three long context types, for rigorous evaluation. Our analysis of eleven LLMs reveals the optimal choice between RAG and LC depends on a complex interplay of model capabilities, context length, task type, and retrieval characteristics, offering actionable guidelines for practitioners. Our code and dataset is provided at:https://github.com/Alibaba-NLP/LaRA

This paper presents a novel and integrated approach for unmanned surface vehicles (USVs) in path planning and trajectory tracking control applications. In order to ensure the optimality and real-time performance of USV’s path planning, the planning algorithm based on reinforcement learning is integrated and improved by fusing the state information of USVs and the environment. To guarantee the smoothness of the generated path curve for the required tracking and control algorithm, the points with large curvature are selected as key points for sampling and quintic spline curve fitting is used. According to the characteristics of USVs during trajectory tracking, a controller is designed to ensure the stability and accuracy of the tracking, and the stability of this controller is proved by Lyapunov function. The validity and feasibility of the proposed method have been verified through simulation experiments.

Eye-tracking is a reliable method for quantifying visual information processing and holds significant potential for group recognition, such as identifying autism spectrum disorder (ASD). However, eye-tracking research typically faces the heterogeneity of stimuli and is time-consuming due to the large number of observed stimuli. To address these issues, we first mathematically define the stimulus selection problem and introduce the concept of stimulus discrimination ability to reduce the computational complexity of the solution. Then, we construct a scanpath-based recognition model to mine the stimulus discrimination ability. Specifically, we propose cross-subject entropy and cross-subject divergence scores for quantitatively evaluating stimulus discrimination ability, effectively capturing differences in intra-group collective trends and inter-subject consistency within a group. Furthermore, we propose an iterative learning mechanism that employs stimulus-wise attention to focus on discriminative stimuli for discrimination purification. In the experiment, we construct an ASD eye-tracking dataset with diverse stimulus types and conduct extensive tests on three representative models to validate our approach. Remarkably, our method demonstrates superior performance using only 10 selected stimuli compared to models utilizing 220 stimuli. Additionally, we perform experiments on another eye-tracking task, gender prediction, to further validate our method. We believe that our approach is both simple and flexible for integration into existing models, promoting large-scale ASD screening and extending to other eye-tracking research domains.

Recognizing autism spectrum disorder (ASD) has faced great challenges due to insufficient professional clinicians and complex procedures. Automated data-driven ASD recognition models can reduce the subjectivity and physician dependency of traditional evaluation methods. Facial data, which can encode important perceptual and social behaviors, have emerged in ASD research to explore novel biomarkers for screening, diagnosing, and treating ASD. However, existing research mainly focuses on extracting low-level hand-crafted facial features for analysis and classification. Determining how to learn discriminative deep representations from dynamic facial data for computational model construction remains an unresolved challenge. In this study, we propose an ASD recognition model based on facial videos to fill the lack of temporal correlation learning of facial features. First, we utilize a vision transformer to extract frame-based global facial features. Then, we use a Longformer to establish the correlation of facial features over time. In the experiment, we recruited 146 subjects between 2 and 8 years of age to record their facial videos under a computer-based eye-tracking experiment and 76 subjects to conduct a smartphone-based experiment. Quantitative comparisons have shown the effectiveness and reliability of the proposed model. Furthermore, we have confirmed the correlation between facial and eye-tracking modalities in visual attention.

In the field of biomedicine, Document-level relation extraction (DocRE) aims to reason about complex relational facts among entities by reading, inferring, and aggregating among entities over multiple sentences in a document. Existing studies construct document-level graphs to enrich interactions between entities. However, these methods pay more attention to the entity nodes and their connections, regardless of the rich knowledge entailed in the original corpus. In this paper, we propose a contrastive learning enhanced document-level graph relation representation(CGDRE) which mines the semantic knowledge from the original corpus and improve the ability of DocRE. Firstly, we use a coreference contrastive learning module to capture the potential semantic knowledge. Secondly, we construct a heterogeneous graph to enhance the graph structure information according to the original document and semantic knowledge. Lastly, CGDRE infers relations on the aggregated graph and uses focal loss to train the model. Remarkably, it is amazing that CGDRE can effectively alleviate the long-tailed distribution problem in the DocRE. Experiments on the public datasets, CDR, GDA and DocRED, show that CGDRE can significantly outperform other baselines, achieving a significant performance improvement. Extensive analyses demonstrate that the performance of our CGDRE is contributed by the capture of the semantic knowledge enhanced graph relation representation.

Document-level relation extraction is an important task in natural language processing that involves identifying and classifying relations between entities mentioned in a document. Traditional approaches often focus on individual sentences or local context, overlooking the broader context of the entire document. In this paper, we propose an Evidence Sentence Augmented Sequence-to-Sequence method for Document-level Relation Extraction(called ESASS-DRE). Our method introduces evidence sentences into the sequence-to-sequence framework to improve document-level relation extraction. The approach consists of two main steps: evidence sentence selection and relation extraction. Firstly, we identify a set of evidence sentences that contain crucial information relevant to the target relation. These sentences are selected based on their importance and contextual relevance. Secondly, the selected evidence sentences are combined with the original document and used as input to the sequence-to-sequence model. These generated sequences are decoded into relation labels, indicating the type of relationship between the entities. By incorporating evidence sentences into the model, we provide additional context and relevant information, enabling the model to make more informed predictions. Experiments conducted on benchmark datasets demonstrate the effectiveness of our method. Compared to traditional approaches, our method achieves higher accuracy and robustness in document-level relation extraction tasks. The incorporation of evidence sentences allows the model to capture the broader context of the document, leading to improved performance. (e.g., by 2.96/3.64 Ign F1/F1 on DocRED).

In the current clinical healthcare environment, there is often a challenge where there is a wealth of unlabeled medical images, but a shortage of labeled images specific to particular medical cases. This limitation restricts the improvement of training effectiveness for deep learning models. This paper, starting from self-supervised learning and drawing inspiration from multi-task learning, explores a Vision Transformer-based self-supervised auxiliary task using MAE for medical image classification. This model establishes both a self-supervised auxiliary task and a supervised primary task, using a shared-weight VIT (Vision Transformer) encoder and synchronously updating network parameters. It effectively leverages both unlabeled and labeled medical images, resulting in promising outcomes. The model is alternately tested on two different medical image classification primary task datasets using three different types of self-supervised auxiliary task sets, and in most cases, it outperforms pure VIT models trained using both supervised and self-supervised learning methods under similar conditions.

The manual adjustment of compilation flags by compiler users is impractical due to the exponential size of the search space. To address this, machine learning-based compiler auto-tuning methods, particularly evolutionary algorithms, have been proposed. However, existing works use different benchmarks and experimental setups, making it difficult to compare the strengths and weaknesses of various algorithms. To address this, we present EAtuner, an evolutionary algorithm-based framework for compiler auto-tuning, with the goal of benchmarking and identifying suitable algorithms for compiler auto-tuning. We implement ten discrete binary evolutionary algorithms and evaluate their effectiveness on the LLVM compiler through experiments. Notably, eight of these algorithms have not been previously applied to compiler flag optimization problems before our work. The results show that all ten algorithms can effectively achieve compiler auto-tuning, resulting in an average speedup of 1.204. However, there are notable differences in the effectiveness and efficiency of each algorithm, particularly in optimization efficiency, which is positively correlated with the number of program compilations. Based on this, we classify the algorithms into three levels, with Differential Evolution (DE) showing significant advantages in optimization effectiveness and efficiency. Additionally, we provide a comprehensive summary of the applicability of compiler flags, the correlation between them, and their relationship with programs.

Current defenses against graph attacks often rely on certain properties to eliminate structural perturbations by identifying adversarial edges from normal edges. However, this dependence makes defenses vulnerable to adaptive (white-box) attacks from adversaries with the same knowledge. Adversarial training seems to be a feasible way to enhance robustness without reliance on artificially designed properties. However, in this paper, we show that it can lead to models learning incorrect information. To solve this issue, we re-examine graph attacks from the out-of-distribution (OOD) perspective for poisoning and evasion attacks and introduce a novel adversarial training paradigm incorporating OOD detection. This approach strengthens the robustness of Graph Neural Networks (GNNs) without reliance on prior knowledge. To further evaluate adaptive robustness, we develop adaptive attacks against our methods, revealing a trade-off between graph attack efficacy and defensibility. Through extensive experiments over 25,000 perturbed graphs, our method could still maintain good robustness against both adaptive and non-adaptive attacks. The code is provided at https://github.com/likuanppd/GOOD-AT.

We propose a novel ensemble learning method based on sample complementarity that considers sample complementarity through scientific model selection and weight allocation. Our approach aims to capture the essence of ensemble learning, which is to achieve complementary advantages between models. Our ensemble method shares a similar level of simplicity in operation as averaging ensemble learning methods, while demonstrating better performance. We first select the model with the best performance to participate in the ensemble. Then, we identify and filter samples that show poor performance in the best model according to the performance to form a subset. We select models with strong complementarity for ensemble learning by evaluating the performance metrics of other models on this subset to determine their level of complementarity with the best model. Furthermore, we reasonably assign ensemble weights by combining the performance of the best model and the sample complementarity of other models. The experimental results show that our proposed method outperforms the averaging ensemble learning methods on the DFUC2022 dataset, the Bank Marketing dataset, and the Predicting Facebook Comment Volume dataset, demonstrating its effectiveness. Our approach achieves 0.7352 mean Dice in the Diabetic Foot Ulcer Challenge 2022 on MICCAI 2022, ranking second on the Live Leaderboard. Our code will be released at https://github.com/xupin262/complementarity.

This paper presents the parameter identification based on generalized orthonormal basis function (GOBF) without persistent excitation via two learning-based paradigms. The problem is formulated as a two-stage identification of poles and weight coefficients of GOBFs. The GOBF is designed based on the two-parameter Kautz basis function for the compatiblity with complex poles. The two learning-based paradigms are established on the nature-inspired meta-heuristic optimization and deep reinforcement learning, respectively. The ability of identification without strict persistent excitation is discussed as well. The effectiveness of the two paradigms are verified through a simulation study on a 4-order transfer function model.

Aiming at the problem of colored noise in the signal, this article proposes a subspace frequency estimation approach under colored noise with application to fault diagnosis of motor rolling bearings. First, a nonlinear discrete-time system is described to generate colored noise. An extended I/O model with parameters of a nonlinear discrete-time system is given by the subspace method. Then, the gap metric-aided system order determination approach is developed for extended observability matrix identification. Then, the data-driven diagnostic observer parameter identification approach and the fast approximate power iterative subspace method are adopted to realize online monitoring for frequency change detection. Eventually, a data-driven design scheme of residual generator is proposed for the implementation of fault detection. The effectiveness of the proposed methods is verified for fault diagnosis performance through numerical simulations and the experimental measurements from the dynamic motor rolling bearing experiment rig.

Atypical eye movement is one of the critical symptoms of autism spectrum disorder (ASD). Automatic quantification of eye-tracking data can provide an objective, convenient, and non-invasive way to identify subjects with ASD, which can develop scalable screening tools for ASD to apply in areas with limited medical resources. However, existing eye-tracking-based ASD classification models usually calculated the score under each image separately and averaged the scores under different images in a post-processing manner to achieve ASD recognition. Determining how to utilize all eye-tracking data of each subject to globally integrate perceptual information and establish a subject-based visual preference for ASD screening is still an unresolved challenge. To address this issue, we propose a novel ASD screening model based on global visual preference encoding. First, we utilize the segment anything model (SAM) and vision transformer (ViT) to extract semantic label regions from all test images. Then, we establish a personalized visual preference graph for each subject based on the saccadic shifts between different semantic regions. Finally, we apply a graph convolutional network (GCN) to learn the mapping between the visual preference graph and classification labels for ASD recognition. In the experiment, we recruited 28 children with ASD and 30 typically developing (TD) children between 2 and 8 years of age to record their eye-tracking data under 220 test images from four types. The experimental results have shown that the proposed model can outperform the state-of-the-art eye-tracking-based ASD recognition models. Furthermore, the evaluation results have indicated the potential to extend the proposed model to other eye-tracking applications, resulting in progress in visual research, accessibility, and healthcare.

Graph Neural Networks (GNNs) have achieved remarkable success in various domains but most of them are developed under the in-distribution assumption. Under out-of-distribution (OOD) settings, they suffer from the distribution shift between the training set and the test set and may not generalize well to the test distribution. Several methods have tried the invariance principle to improve the generalization of GNNs in OOD settings. However, in previous solutions, the graph encoder is immutable after the invariant learning and cannot be adapted to the target distribution flexibly. Confronting the distribution shift, a flexible encoder with refinement to the target distribution can generalize better on the test set than the stable invariant encoder. To remedy these weaknesses, we propose a Flexible invariant Learning framework for Out-Of-Distribution generalization on graphs (FLOOD), which comprises two key components, invariant learning and bootstrapped learning. The invariant learning component constructs multiple environments from graph data augmentation and learns invariant representation under risk extrapolation. Besides, the bootstrapped learning component is devised to be trained in a self-supervised way with a shared graph encoder with the invariant learning part. During the test phase, the shared encoder is flexible to be refined with the bootstrapped learning on the test set. Extensive experiments are conducted for both transductive and inductive node classification tasks. The results demonstrate that FLOOD consistently outperforms other graph OOD generalization methods and effectively improves the generalization ability.

With the development of Cyber-Physical Systems (CPSs), many industrial facilities have realized remote control and monitoring. However, the widespread of CPSs has brought new issues and challenges in terms of security. Attackers can exploit vulnerabilities induced by network communication, tamper with transmitted data, and cause serious accidents through carefully designed covert attacks. This paper proposes a residual-driven comprehensive defense scheme based on the coprime factorization technique to address the threat posed by concealed CPS attacks. The novel scheme protects CPS from stealth cyber-physical attacks through secure transmission and attack detection. In particular, a secure transmission method is first introduced to prevent information leakage from the source. The pivotal idea is to convert confidential transmission control and measurement signals into non-essential filtered residual signals. It contributes to the reduction of information leakage and helps reduce the risks of stealth attacks. Then, under the same residual-driven framework, a stealth attack detection approach is put forward. It can eliminate false alarms caused by system faults, and therefore, achieve superior efficacy in detection accuracy under stealth attacks. Finally, simulation research is conducted on the F-404 engine to verify the effectiveness and performance of the proposed scheme and approach.

CD56, one of the latest tumor molecular markers, is a nerve cell adhesion molecule that can be used in the diagnosis and study of a variety of tumor cells. In the field of digital medical image processing, research on CD56 images is emerging. In CD56 images, cells are densely packed and the number of positive cells is low, these make it difficult to accurately detect cells in CD56 images. In this paper, we proposed a new hierarchy method to detect positive cells from CD56 images. Specifically, we first utilize an ExtremeNet to roughly detect positive and negative cells from an image. Then Shape Constraints are adopted to refine the detection results. We further trained a convolutional neural network to identify negative and positive cells more accurately, which is called Re-Discrimination. Experimental results showed that compared with several state-of-the-art one-stage object detection algorithms, the proposed method achieved the highest detection accuracy of cells in CD56 images.

Benefiting from the message passing mechanism, Graph Neural Networks (GNNs) have been successful on flourish tasks over graph data. However, recent studies have shown that attackers can catastrophically degrade the performance of GNNs by maliciously modifying the graph structure. A straightforward solution to remedy this issue is to model the edge weights by learning a metric function between pairwise representations of two end nodes, which attempts to assign low weights to adversarial edges. The existing methods use either raw features or representations learned by supervised GNNs to model the edge weights. However, both strategies are faced with some immediate problems: raw features cannot represent various properties of nodes (e.g., structure information), and representations learned by supervised GNN may suffer from the poor performance of the classifier on the poisoned graph. We need representations that carry both feature information and as mush correct structure information as possible and are insensitive to structural perturbations. To this end, we propose an unsupervised pipeline, named STABLE, to optimize the graph structure. Finally, we input the well-refined graph into a downstream classifier. For this part, we design an advanced GCN that significantly enhances the robustness of vanilla GCN [24] without increasing the time complexity. Extensive experiments on four real-world graph benchmarks demonstrate that STABLE outperforms the state-of-the-art methods and successfully defends against various attacks.

Though Graph Neural Networks (GNNs) have been successful for fraud detection tasks, they suffer from imbalanced labels due to limited fraud compared to the overall userbase. This paper attempts to resolve this label-imbalance problem for GNNs by maximizing the AUC (Area Under ROC Curve) metric since it is unbiased with label distribution. However, maximizing AUC on GNN for fraud detection tasks is intractable due to the potential polluted topological structure caused by intentional noisy edges generated by fraudsters. To alleviate this problem, we propose to decouple the AUC maximization process on GNN into a classifier parameter searching and an edge pruning policy searching, respectively. We propose a model named AO-GNN (Short for AUC-oriented GNN), to achieve AUC maximization on GNN under the aforementioned framework. In the proposed model, an AUC-oriented stochastic gradient is applied for classifier parameter searching, and an AUC-oriented reinforcement learning module supervised by a surrogate reward of AUC is devised for edge pruning policy searching. Experiments on three real-world datasets demonstrate that the proposed AO-GNN patently outperforms state-of-the-art baselines in not only AUC but also other general metrics, e.g. F1-macro, G-means.

In online learning, the dynamic regret metric chooses the reference oracle that may change over time, while the typical (static) regret metric assumes the reference solution to be constant over the whole time horizon. The dynamic regret metric is particularly interesting for applications, such as online recommendation (since the customers' preference always evolves over time). While the online gradient (OG) method has been shown to be optimal for the static regret metric, the optimal algorithm for the dynamic regret remains unknown. In this article, we show that proximal OG (a general version of OG) is optimum to the dynamic regret by showing that the proved lower bound matches the upper bound. It is highlighted that we provide a new and general lower bound of dynamic regret. It provides new understanding about the difficulty to follow the dynamics in the online setting.

In the field of human-computer interaction, video-based gesture detection has already sparked a lot of attention. Unlike typical RGB gesture movies, RGB-D gesture videos also record the depth information corresponding to all of the pixels in each frame, potentially reducing the impact of lighting and background fluctuations. To the best of our knowledge, current RGB-D gesture video datasets primarily focus on high classification accuracy while ignoring adequate illumination and backdrop variations that occur in real life. These will stifle the advancement of gesture recognition algorithms in some way. Instead, we present DG-13, an RGB-D gesture video dataset that properly accounts for various brightness and backdrops. On the proposed DG-13 dataset, benchmark evaluations of five exemplary light-weighted 3D CNN networks are also provided. Experiment results reveal that when there are significant illumination and backdrop differences, RGB-D gesture video can successfully aid increase classification accuracy and numerous other classification metrics. The DG-13 dataset and benchmark codes are public at https://github.com/xiaooo-jian/DG-13.

Cup anemometers are widely used instruments for wind turbines to measure wind speed in wind farm. Aimed to reduce the adverse impact on wind energy resource estimation, this paper proposes a data-driven fault diagnosis approach for assessing the anemometer health status. Auto-associative netural network (AANN) is developed to reconstruct the anemometer measurement data after data pre-processing, and residual analysis is performed between the anemometer measurement data and the AANN reconstruction data. In addition, the quantitative indicators that can reflect the health status of the anemometer gained from residuals are obtained through the K-Means clustering algorithm, based on which the faulty anemometers in the wind farm can be identified. The approach can provide guidance for the production and operation of the wind farm.

A data-driven disturbance decoupling FTC for system with deterministic disturbance is proposed in this paper. The algorithm of subspace identification and modified Partial least square is aided to improve the limit of dynamic linearization based predictive control. Compared with the existing dynamic linearization based predictive control, the proposed control strategy enhances its control performance by reducing its sensitivity to noise and decoupling the disturbance. The efficiency of the proposed FTC approach compared to the ynamic linearization based predictive controller is suggested in the simulation of a DC motor.

Industrial systems are always subjected to the deterministic disturbance due to some inherited factors, which is likely to degrade the control and monitoring performance to some extent. This paper presents an approach to the closed-loop subspace identification of the data-driven stable kernel representation (SKR) with the deterministic disturbance. The essence is that we extend the CSIMPCA algorithm by introducing the deterministic disturbance and subsequently separate the part corresponding to the SKR of the system from the obtained parity space. The inspiration for the idea mainly stems from the necessity for the identification and process monitoring of practical closed-loop systems. The effectiveness of the proposed method is demonstrated and illustrated through randomly generated 4-order MIMO discrete-time LTI systems. Furthermore, the identified SKR is finally applied to the fault detection and related experimental results show a decent detection performance.

This paper proposes a new method to estimate the stability margin of a system by factorizing it into system's data-driven stable kernel representation (SKR) and controller's stable image representation (SIR). To this end, a coprime factorization technology is applied to the closed-loop system firstly. By analyzing the relations between the reference signal and internal signal, a new approach is adopted to calculate SKR of the system by the least square (LS) method. Furthermore, the data-driven realization of stability margin is calculated through system's SKR and controller's SIR. An example is given in the last part to testify the correctness of methodologies proposed in this paper.

Stochastic gradient descent (SGD) is a commonly used technique in large-scale machine learning tasks, but its convergence is slow due to the inherent variance. In recent years, a popular method, Stochastic Variance Reduced Gradient (SVRG), addresses this shortcoming via computing the full gradient of the entire dataset in each epoch. However, conventional SVRG and its variants usually need to identify a hyperparameter - the epoch size, which is essential to the convergence performance. Few previous studies discuss how to systematically find a suitable value for that hyper-parameter, which makes it hard to gain a good convergence performance in practical machine learning applications. In this paper, we propose a new stochastic gradient descent named AESVRG, which introduces variance reduction and computes the full gradient adaptively. Its enhanced implementation, AESVRG+, has a convergence performance that can outplay existing SVRG with fine-tuned epoch sizes. An extensive evaluation illustrates the significant performance improvement of our method.

Heterotrimeric G protein signaling cascades are one of the primary metazoan sensing mechanisms linking a cell to environment. However, the number of experimentally identified effectors of G protein in plant is limited. We have therefore studied which tools are best suited for predicting G protein effectors in rice. Here, we compared the predicting performance of four classifiers with eight different encoding schemes on the effectors of G proteins by using 10-fold cross-validation. Four methods were evaluated: random forest, naive Bayes, K-nearest neighbors and support vector machine. We applied these methods to experimentally identified effectors of G proteins and randomly selected non-effector proteins, and tested their sensitivity and specificity. The result showed that random forest classifier with composition of K-spaced amino acid pairs and composition of motif or domain (CKSAAP_PROSITE_200) combination method yielded the best performance, with accuracy and the Mathew's correlation coefficient reaching 74.62% and 0.49, respectively. We have developed G-Effector, an online predictor, which outperforms BLAST, PSI-BLAST and HMMER on predicting the effectors of G proteins. This provided valuable guidance for the researchers to select classifiers combined with different feature selection encoding schemes. We used G-Effector to screen the effectors of G protein in rice, and confirmed the candidate effectors by gene co-expression data. Interestingly, one of the top 15 candidates, which did not appear in the training data set, was validated in a previous research work. Therefore, the candidate effectors list in this article provides both a clue for researchers as to their function and a framework of validation for future experimental work. It is accessible at http://bioinformatics.fafu.edu.cn/geffector.

Automatic identification of abnormal cervical cells, including feature representation, feature combination and classification strategy, is highly demanded in women's annual cervical cancer screenings. However, previous methods only deal with one or two of these three phases, and currently there is few complete framework for this problem. A novel three‐phrase boosting framework is proposed for the detection of abnormal cells from cervical smear images. First, the authors extract 160 dimensional features with respect to each cervical cell from three aspects, including cytology morphology, chromatin pathology and region intensity. In particular, 106 dimensional chromatin pathology features are newly adopted to describe the nucleus textural transformation. Second, an adaptive feature combination method is introduced to select the optimal feature patterns, which can combine all features using a reinforced margin‐based approach with the heuristic knowledge. Finally, a two‐stage classification strategy is presented to reduce erroneous classification abnormal cells using two different classifiers. Experimental results achieve state‐of‐the‐art performance and the proposed framework outperforms the other 16 compared detection methods.

In an abnormal cervical cell detection system the discriminated abilities of different features are not same so the optimized combination method of all features is an essential component to this system. Feature selection can improve each feature utilization ratio and the performance of the classification problem. The previous efforts of cervical abnormal cell detection are mainly focused on changing feature space into a new one by using a binary weight vector. In this work, the binary weight values are extended to the multiple weight values. According to the statistical distribution situation of the data, an adaptive margin-based weighted feature selection method is proposed in this paper. This method performs best compared with the other 3 methods. The experimental result achieves 96% accuracy in a real-world cervical smear image dataset.

This paper presents an implementation of an accurate and efficient compensated Double-precision General Matrix Multiplication (DGEMM) based on OpenBLAS for 64-bit ARMv8 multi-core processors. Due to cancellation phenomena in floating point arithmetic, the results of DGEMM may not be as accurate as expected. In order to increase the accuracy of DGEMM, we compensate the error introduced by its dot product kernel (GEBP) by applying an error-free transformation to rewrite the kernel in assembly language. We optimize the computations in the inner kernel through exploiting loop unrolling, instruction scheduling and software-implemented register rotation to exploit instruction level parallelism (ILP). We also conduct a priori error analysis of the derived CompDGEMM. Our compensated DGEMM is as accurate as the existing quadruple precision GEMM using MBLAS, but is up to 6.4x faster. Our parallel implementation achieves good performance and scalability under varying thread counts across a range of matrix sizes evaluated.

This paper proposes a new non-reference image quality metric that can be adopted by the state-of-the-art image/ video denoising algorithms for auto-denoising. The proposed metric is extremely simple and can be implemented in four lines of Matlab code. The basic assumption employed by the proposed metric is that the noise should be independent of the original image. A direct measurement of this dependence is, however, impractical due to the relatively low accuracy of existing denoising method. The proposed metric thus aims at maximizing the structure similarity between the input noisy image and the estimated image noise around homogeneous regions and the structure similarity between the input noisy image and the denoised image around highly-structured regions, and is computed as the linear correlation coefficient of the two corresponding structure similarity maps. Numerous experimental results demonstrate that the proposed metric not only outperforms the current state-of-the-art non-reference quality metric quantitatively and qualitatively, but also better maintains temporal coherence when used for video denoising.

Semantic Role Labeling (SRL) for tweets is a meaningful task that can benefit a wide range of applications such as fine-grained information extraction and retrieval from tweets. One main challenge of the task is the lack of annotated tweets, which is required to train a statistical model. We introduce self-training to SRL, leveraging abundant unlabeled tweets to alleviate its depending on annotated tweets. A novel strategy of tweet selection is presented, ensuring the chosen tweets are both correct and informative. More specifically, the correctness is estimated according to the labeling confidences and agreement of two Conditional Random Fields based labelers, which are trained on the randomly evenly spitted labeled data; while the informativeness is in proportion to the maximum distance between the tweet and the already selected tweets. We evaluate our method on a human annotated data set and show that bootstrapping improve a baseline by 3.4% F1.

As tweets have become a comprehensive repository of fresh information, Semantic Role Labeling (SRL) for tweets has aroused great research interests because of its central role in a wide range of tweet related studies such as fine-grained information extraction, sentiment analysis and summarization. However, the fact that a tweet is often too short and informal to provide sufficient information poses a major challenge. To tackle this challenge, we propose a new method to collectively label similar tweets. The underlying idea is to exploit similar tweets to make up for the lack of information in a tweet. Specifically, similar tweets are first grouped together by clustering. Then for each cluster a two-stage labeling is conducted: One labeler conducts SRL to get statistical information, such as the predicate/argument/role triples that occur frequently, from its highly confidently labeled results; then in the second stage, another labeler performs SRL with such statistical information to refine the results. Experimental results on a human annotated dataset show that our approach remarkably improves SRL by 3.1 % F1.