基于RNN汉语语言模型自适应算法研究

深度学习在自然语言处理中的应用越来越广泛。相比于传统的n-gram统计语言模型,循环神经网络(Recurrent Neural Network,RNN)建模技术在语言模型建模方面表现出了极大的优越性,逐渐在语音识别、机器翻译等领域中得到应用。然而,目前RNN语言模型的训练大多是离线的,对于不同的语音识别任务,训练语料与识别任务之间存在着语言差异,使语音识别系统的识别率受到影响。在采用RNN建模技术训练汉语语言模型的同时,提出一种在线RNN模型自适应(self-adaption)算法,将语音信号初步识别结果作为语料继续训练模型,使自适应后的RNN模型与识别任务之间获得最大程度的匹配。实验结果表明:自适应模型有效地减少了语言模型与识别任务之间的语言差异,对汉语词混淆网络进行重打分后,系统识别率得到进一步提升,并在实际汉语语音识别系统中得到了验证。     

嵌入式在线航空摄影测量方法

为满足空间数据实时处理的需求,提高航空摄影测量系统在线处理能力,研究一种在线摄影测量的理论与方法,运用嵌入式架构设计一种在线摄影测量系统,使用可编程门阵列(FPGA)方法、高级精简指令集机器组合数字信号处理器(ARM+DSP)方法等嵌入式计算技术建立专门的硬件运行环境,移植并优化现有算法到嵌入式系统中,实现摄影测量数据的在线处理。机载航摄实验结果表明,该方法能够实现对摄影测量数据进行快速稳定的在线处理,验证了运用嵌入式架构的在线摄影测量的可行性,把摄影测量的处理效率提高到了一个更高的水平,同时也进一步拓宽了摄影测量技术的应用领域。     

基于在线动平衡的磁轴承参数辨识

针对磁悬浮刚性转子系统,提出了一种基于在线动平衡的磁轴承参数辨识方法。采用已知大小和相位的不平衡质量作为激振源,根据动平衡仪测得的转子不平衡响应,求得磁轴承位移刚度和电流刚度。利用某型磁悬浮控制力矩陀螺对该方法进行了试验验证,得到了一定转速范围内的磁轴承位移刚度和电流刚度,证明了该方法的正确性和实用性。试验结果表明外转子型磁轴承位移刚度和电流刚度随转速的增加而下降明显。     

Improved bounds for batch scheduling with nonidentical job sizes

Here, we revisit the bounded batch scheduling problem with nonidentical job sizes on single and parallel identical machines, with the objective of minimizing the makespan. For the single machine case, we present an algorithm which calls an online algorithm (chosen arbitrarily) for the one‐dimensional bin‐packing problem as a sub‐procedure, and prove that its worst‐case ratio is the same as the absolute performance ratio of . Hence, there exists an algorithm with worst‐case ratio , which is better than any known upper bound on this problem. For the parallel machines case, we prove that there does not exist any polynomial‐time algorithm with worst‐case ratio smaller than 2 unless P = NP, even if all jobs have unit processing time. Then we present an algorithm with worst‐case ratio arbitrarily close to 2. © 2014 Wiley Periodicals, Inc. Naval Research Logistics 61: 351–358, 2014     

On competitive analysis for polling systems

Polling systems have been widely studied, however most of these studies focus on polling systems with renewal processes for arrivals and random variables for service times. There is a need driven by practical applications to study polling systems with arbitrary arrivals (not restricted to time-varying or in batches) and revealed service time upon a job's arrival. To address that need, our work considers a polling system with generic setting and for the first time provides the worst-case analysis for online scheduling policies in this system. We provide conditions for the existence of constant competitive ratios, and competitive lower bounds for general scheduling policies in polling systems. Our work also bridges the queueing and scheduling communities by proving the competitive ratios for several well-studied policies in the queueing literature, such as cyclic policies with exhaustive, gated or l-limited service disciplines for polling systems.     

环境参数在线辨识及其在滑翔段制导中的应用

针对助推滑翔导弹,提出基于飞行环境参数在线辨识的滑翔段数值预测校正制导方法。综合考虑大气密度和气动系数对导弹运动的影响,引入滑翔段综合环境参数,利用扩展Kalman滤波方法对综合参数进行在线辨识。基于在线辨识结果,利用渐消记忆递推最小二乘方法在线建立环境参数预测模型,并利用最新辨识结果进行模型的在线修正。设计了纵向和横向制导律,并基于环境参数在线预测模型进行落点预测,以克服飞行环境扰动对落点预测精度的影响。进行了大气密度非定常扰动下的制导仿真,以及密度和气动参数随机扰动下的Monte Carlo仿真。仿真结果表明:环境参数在线预测模型能准确预报飞行环境参数,制导方法对飞行环境扰动具有较强的鲁棒性。     

Optimal inventory and dynamic admission policies for a retailer of seasonal products with affiliate programs and drop‐shipping

This article investigates the optimal inventory and admission policies for a “Clicks‐and‐Bricks” retailer of seasonal products that, in addition to selling through its own physical and online stores, also sells through third‐party websites by means of affiliate programs. Through postings on partners' webpages, an affiliate program allows a retailer to attract customers who would otherwise be missed. However, this retailer needs to pay a commission for each sale that originates from the website operators participating in the program. The retailer may also refer online orders to other sources (such as distributors and manufacturers) for fulfillment through a drop‐shipping agreement and thus earns commissions. This would be an option when, for example, the inventories at the physical stores were running low. Therefore, during the selling horizon, the retailer needs to dynamically control the opening/closing of affiliate programs and decide on the fulfillment option for online orders. On the basis of a discrete‐time dynamic programming model, the optimal admission policy of the retailer is investigated in this paper, and the structural properties of the revenue function are characterized. Numerical examples are given to show the revenue impact of optimal admission control. The optimal initial stocking decisions at the physical stores are also studied. © 2009 Wiley Periodicals, Inc. Naval Research Logistics 2009     

On online bin packing with LIB constraints

In many applications of packing, the location of small items below large items, inside the packed boxes, is forbidden. We consider a variant of the classic online one‐dimensional bin packing, in which items allocated to each bin are packed there in the order of arrival, satisfying the condition above. This variant is called online bin packing problem with LIB (larger item in the bottom) constraints. We give an improved analysis of First Fit showing that its competitive ratio is at most , and design a lower bound of 2 on the competitive ratio of any online algorithm. In addition, we study the competitive ratio of First Fit as a function of an upper bound (where d is a positive integer) on the item sizes. Our upper bound on the competitive ratio of First Fit tends to 2 as d grows, whereas the lower bound of two holds for any value of d. Finally, we consider several natural and well known algorithms, namely, Best Fit, Worst Fit, Almost Worst Fit, and Harmonic, and show that none of them has a finite competitive ratio for the problem. © 2009 Wiley Periodicals, Inc. Naval Research Logistics, 2009     

控制性学习条件下网络外语教学环境创建刍议

网络教学环境的创建及其在外语教学中的运用越来越成为外语界关注的焦点。在控制环境下建构多媒体网络教学环境的手段,充分调动学生的主动性和参与性,从而有助于提高学习者的语言水平及语用能力,加速其文化及社会知识的建构。     

Doubly Stochastic Sequential Assignment Problem

This article introduces the Doubly Stochastic Sequential Assignment Problem (DSSAP), an extension of the Sequential Stochastic Assignment Problem (SSAP), where sequentially arriving tasks are assigned to workers with random success rates. A given number of tasks arrive sequentially, each with a random value coming from a known distribution. On a task arrival, it must be assigned to one of the available workers, each with a random success rate coming from a known distribution. Optimal assignment policies are proposed for DSSAP under various assumptions on the random success rates. The optimal assignment algorithm for the general case of DSSAP, where workers have distinct success rate distribution, has an exponential running time. An approximation algorithm that achieves a fraction of the maximum total expected reward in a polynomial time is proposed. The results are illustrated by several numerical experiments. © 2016 Wiley Periodicals, Inc. Naval Research Logistics 63: 124–137, 2016