Deinterleaving pulse trains using discrete-time stochasticdynamic-linear models

Pulse trains from a number of different sources are often received on the one communication channel. It is then of interest to identify which pulses are from which source, based on different source characteristics. This sorting task is termed deinterleaving. the authors propose time-domain techniques for deinterleaving pulse trains from a finite number of periodic sources based on the time of arrival (TOA) and pulse energy, if available, of the pulses received on the one communication channel. They formulate the pulse train deinterleaving problem as a stochastic discrete-time dynamic linear model (DLM), the “discrete-time” variable k being associated with the kth received pulse. The time-varying parameters of the DLM depend on the sequence of active sources. The deinterleaving detection/estimation task can then be done optimally via linear signal processing using the Kalman filter (or recursive least squares when the source periods are constant) and tree searching. The optimal solution, however, is computationally infeasible for other than small data lengths since the number of possible sequences grow exponentially with data length. The authors propose and study two of a number of possible suboptimal solutions: 1) forward dynamic programming with fixed look-ahead rather than total look-ahead as required for the optimal scheme; 2) a probabilistic teacher Kalman filtering for the detection/estimation task     

Classification‐Based Approach for Hybridizing Statistical and Rule‐Based Machine Translation

In this paper, we propose a classification‐based approach for hybridizing statistical machine translation and rule‐based machine translation. Both the training dataset used in the learning of our proposed classifier and our feature extraction method affect the hybridization quality. To create one such training dataset, a previous approach used auto‐evaluation metrics to determine from a set of component machine translation (MT) systems which gave the more accurate translation (by a comparative method). Once this had been determined, the most accurate translation was then labelled in such a way so as to indicate the MT system from which it came. In this previous approach, when the metric evaluation scores were low, there existed a high level of uncertainty as to which of the component MT systems was actually producing the better translation. To relax such uncertainty or error in classification, we propose an alternative approach to such labeling; that is, a cut‐off method. In our experiments, using the aforementioned cut‐off method in our proposed classifier, we managed to achieve a translation accuracy of 81.5% — a 5.0% improvement over existing methods.     

一种雷达信号处理多任务调度算法

雷达信号处理(RSP)系统的实时性一直是系统设计者需要重点考虑的内容之一。为提高雷达系统的实时性,本文提出了一种基于多模式雷达的RSP任务模型,并根据DP-Wrap算法和流水处理的思想提出了一种高效的任务调度算法。研究了在本模型下影响系统最小处理时间的因素,给出了系统最小处理时间的精确表达式,并在此基础上进行数值仿真,给出了根据RSP任务选择处理器个数的依据。     

Zero-anaphora resolution in Korean based on deep language representation model: BERT

It is necessary to achieve high performance in the task of zero anaphora resolution (ZAR) for completely understanding the texts in Korean, Japanese, Chinese, and various other languages. Deep-learning-based models are being employed for building ZAR systems, owing to the success of deep learning in the recent years. However, the objective of building a high-quality ZAR system is far from being achieved even using these models. To enhance the current ZAR techniques, we fine-tuned a pre-trained bidirectional encoder representations from transformers (BERT). Notably, BERT is a general language representation model that enables systems to utilize deep bidirectional contextual information in a natural language text. It extensively exploits the attention mechanism based upon the sequence-transduction model Transformer. In our model, classification is simultaneously performed for all the words in the input word sequence to decide whether each word can be an antecedent. We seek end-to-end learning by disallowing any use of hand-crafted or dependency-parsing features. Experimental results show that compared with other models, our approach can significantly improve the performance of ZAR.     

利用配价信息的语义角色标注

语义角色标注是一种浅层语义分析.现有的汉语语义分析方法和语义角色标注体系没有结合汉语的特点并有效刻画出汉语的本质特性,导致目前汉语语义角色标注性能与英语相比相差较大.在汉语中,配价结构可以较好地刻画汉语句子的句法结构和语义构成关系,因此,我们在考察配价语法的基础上适当修改了语义角色标注体系并将谓词本身的配价信息融入语义角色标注.实验结果表明,配价信息的使用能够较大幅度提高动名词性谓词的语义角色标注性能:基于正确句法树和正确谓词识别,动词性谓词的SRL性能F1值达到93.69％;名词性谓词的SRL性能F1值达到79.23％;均优于目前国内外的同类系统.     

Whither speech recognition: the next 25 years

The fundamentals of speech recognition are reviewed. The dimensions of the speech recognition task, speech feature analysis, pattern classification using hidden Markov models, language processing, and the current accuracy of speech recognition systems are discussed. The applications of speech recognition in telecommunications, voice dictation, speech understanding for data retrieval, and consumer products are examined     

Mass transit power traction networks as communication channels

The rapid progress of power line communications (PLC) for data transmission over electric power supply systems is now opening ways for special applications such as train automation in local transportation and mass transit (MT) systems. These DC-powered traction networks can be used as communication links between wayside equipment and the moving trains. As MT networks significantly differ from usual electricity supply systems, the usage of existing models and communication equipment for conventional PLC channels turns out infeasible. Therefore, the work reported in this paper focuses on MT channel investigation and modeling, in order to develop novel adapted solutions. The outcome is a stochastic MT channel model, which-besides multipath and time-variance-also includes peculiar properties such as the behavior of ring structures and the impact of the Doppler effect invoked by moving trains. In addition, a very special interference scenario is treated, caused by the rectifiers in these DC-powered environments. Besides a complete simulation model, this paper presents detailed guidelines for building emulation hardware, so that channel adapted PLC system design for MT networks can now be successfully started without further expensive field trials.     

A Novel Dual Pointer Approach for Entity Mention Extraction

The named entity extraction task aims to extract entity mentions from the unstructured text, including names of people, places, institutions and so on. It plays an important role in many Natural language processing (NLP) tasks, such as knowledge bases construction, automatic question answering system and information extraction. Most of the existing entity extraction studies are based on the long text data, which are easier to annotate due to the sufficient contextual information. Extracting entities from short texts such as search queries, conversations is still a challenging task. This paper proposes a dual pointer approach for entity mention extraction, it extracts one entities by two position pointers of the input sentence. The end-to-end deep neural networks model based on the proposed approach can extract the entities by serially generating the dual pointers. The evaluation results on the Chinese public dataset show that the model achieves the state-of-the-art results over the baseline models.     

Wavelet domain statistical hyperspectral soil texture classification

This communication presents an automatic soil texture classification system using hyperspectral soil signatures and wavelet-based statistical models. Previous soil texture classification systems are closely related to texture classification methods, where images are used for training and testing. In this study, we develop a novel system using hyperspectral soil textures, which provide rich information and intrinsic properties about soil textures, where two wavelet-domain statistical models, namely, the maximum-likelihood and hidden Markov models, are incorporated for the classification task. Experimental results show that these methods are both reliable and robust.     

Model-guided labeling of coronary structure

Assigning anatomic labels to coronary arteries in X-ray angiograms is an important task in medical imaging, motivated by the desire to standardize the assessment of coronary artery disease and to facilitate the three-dimensional (3-D) reconstruction and visualization of the coronary vasculature. However, automatic labeling poses a number of significant challenges, including the presence of noise, artifacts, competing structures, misleading visual cues, and other difficulties associated with a dynamic and inherently complex structure. The authors have developed a model-guided approach that addresses these challenges and automatically labels the vascular structure in coronary angiographic images. The approach consists of two models: (1) a symbolic model, represented through a directed acyclic graph, that captures vascular tree hierarchies and branch interrelationships and (2) a generalized 3-D model that captures spatial and geometric relationships. Importantly, the approach detects ambiguities (such as vessel overlaps) that may be found in a frame of a cine sequence, and resolves these ambiguities by considering the information derived from other (unambiguous) frames in the temporal sequence, employing dynamic programming methods to match the image features found in the different (ambiguous and unambiguous) frames. This paper presents this model-guided labeling algorithm and discusses the experimental results obtained from implementing and applying the resulting labeling system to a variety of clinical images. The results indicate the feasibility of achieving robust and consistently accurate image labeling through this model-guided, temporal disambiguation method