Results on memory-limited U-shaped learning

U-shaped learning is a learning behaviour in which the learner first learns a given target behaviour, then unlearns it and finally relearns it. Such a behaviour, observed by psychologists, for example, in the learning of past-tenses of English verbs, has been widely discussed among psychologists and cognitive scientists as a fundamental example of the non-monotonicity of learning. Previous theory literature has studied whether or not U-shaped learning, in the context of Gold’s formal model of learning languages from positive data, is necessary for learning some tasks.It is clear that human learning involves memory limitations. In the present paper we consider, then, the question of the necessity of U-shaped learning for some learning models featuring memory limitations. Our results show that the question of the necessity of U-shaped learning in this memory-limited setting depends on delicate tradeoffs between the learner’s ability to remember its own previous conjecture, to store some values in its long-term memory, to make queries about whether or not items occur in previously seen data and on the learner’s choice of hypotheses space.     

U-shaped, iterative, and iterative-with-counter learning

This paper solves an important problem left open in the literature by showing that U-shapes are unnecessary in iterative learning from positive data. A U-shape occurs when a learner first learns, then unlearns, and, finally, relearns, some target concept. Iterative learning is a Gold-style learning model in which each of a learner’s output conjectures depends only upon the learner’s most recent conjecture and input element. Previous results had shown, for example, that U-shapes are unnecessary for explanatory learning, but are necessary for behaviorally correct learning. Work on the aforementioned problem led to the consideration of an iterative-like learning model, in which each of a learner’s conjectures may, in addition, depend upon the number of elements so far presented to the learner. Learners in this new model are strictly more powerful than traditional iterative learners, yet not as powerful as full explanatory learners. Can any class of languages learnable in this new model be learned without U-shapes? For now, this problem is left open.     

U形波纹管疲劳寿命研究

疲劳寿命是U形波纹管的重要指标.利用传统公式和MSC Fatigue软件对波纹管的应力和寿命进行了计算.经试验验证和分析,认为利用MSC Fatigue软件进行疲劳寿命的方法是可行的,为波纹管的寿命分析提供了新的研究途径.     

基于近似U型网络结构的图像去噪模型

针对图像去噪中的去噪效果差、训练周期长的问题,提出一种基于近似U型网络结构的图像去噪模型。首先,使用不同步长的卷积层将原有的线性网络结构修改为近似U型的网络结构;然后,将不同感受野的图像信息叠加以尽可能地保留图像的原有信息;最后,引入反卷积网络层进行图像恢复和噪声的进一步去除。在Set12与BSD68测试集上与去噪卷积神经网络（DnCNN）模型相比,所提模型的峰值信噪比（PSNR）平均提升了0.04~0.14 dB,训练时长平均缩短了41%。实验结果表明,所提模型具有更好地去噪效果和更短的训练时长。     

Balancing of parallel U-shaped assembly lines

A new hybrid assembly line design, called parallel U-shaped assembly line system, is introduced and characterised along with numerical examples for the first time. Different from existing studies on U-shaped lines, we combine the advantages of two individual line configurations (namely parallel lines and U-shaped lines) and create an opportunity for assigning tasks to multi-line workstations located in between two adjacent U-shaped lines with the aim of maximising resource utilisation. Utilisation of crossover workstations, in which tasks from opposite areas of a same U-shaped line can be performed, is also one of the main advantages of the U-shaped lines. As in traditional U-shaped line configurations, the newly proposed line configuration also supports the utilisation of crossover workstations. An efficient heuristic algorithm is developed to find well-balanced solutions for the proposed line configurations. Test cases derived from existing studies and modified in accordance with the proposed system in this study are solved using the proposed heuristic algorithm. The comparison of results obtained when the lines are balanced independently and when the lines are balanced together (in parallel to each other) clearly indicates that the parallelisation of U-shaped lines helps decrease the need for workforce significantly.     

Variations on ray shooting

We solve some problems related toray shooting in the plane, such as finding the first object hit by a query ray or counting the number of objects intersected by the query line. Our main results are an algorithm for finding the first hit when the objects are lines, and an algorithm for the case when the objects are segments. If the segments form simple polygons, this information can be used for reducing the complexity of the algorithms. The algorithms are efficient in space and in query time. Moreover, they are simple and therefore of practical use.This research was partially supported by the NY Metropolitan Research Fund. The second author is currently at IBM Haifa Research Group, Haifa 32000, Israel.     

基于非对称U型卷积神经网络的脑肿瘤图像分割

在计算机视觉分割任务中, 基于Transformer的图像分割模型需要大量的图像数据才能达到最好的性能, 医学图像相对于自然图像, 数据量非常稀少, 而卷积本身具有更高的感应偏差, 使得它更适合医学图像方面的应用. 为了将Transformer的远程表征学习与CNN的感应偏差相结合, 本文设计了残差ConvNeXt模块来模拟Transformer的设计结构, 采用深度卷积和逐点卷积组成的残差ConvNeXt模块来提取特征信息, 极大地降低了参数量. 并对感受野和特征通道进行了有效的缩放和扩展, 丰富了特征信息. 此外, 本文提出了一个非对称3D U型网络ASUNet用于脑肿瘤图像的分割. 在非对称U型结构中, 采用残差连接, 将最后两个编码器的输出特征进行连接来扩大通道数. 最后, 在上采样的过程中采用深度监督, 促进了上采样过程中语义信息的恢复. 在BraTS 2020和FeTS 2021数据集上的实验结果表明, ET、WT和TC的骰子分数分别达到了77.08%、90.83%、83.41%和75.63%、90.45%、84.21%. 并且通过对比实验, ASUNet在准确性方面完全可以与Transformer构建的模型竞争, 同时保持了标准卷积神经网络的简单性和高效性.     

基于U型网络的K-TIG焊熔池图像分割方法研究

小孔钨极惰性气体保护焊K-TIG(Keyhole Tungsten Inert Gas Welding)的强烈氩弧光会对熔池液面与焊件的成像造成干扰。针对其熔池图像呈现边界模糊、特征尺寸差异大、形态不规则的特点,设计了一种基于注意力机制的多尺度特征融合语义分割模型,对焊接熔池区域图像进行分割。首先,利用并行结构的多尺度非对称瓶颈单元替换UNet在特征提取过程中的卷积块,加强对不同尺度特征提取的能力,强化对熔池轮廓的表征能力;然后,在上采样阶段使用跨层的注意力模块引导网络模型更加关注熔池的锁孔区域;最后,在进行网络训练之前,对训练集的图片采用Multi-Scale Retinex算法执行颜色和形态的图像增强。实验结果显示,该神经网络的分割结果与熔池实际区域在Dice系数、MIOU以及F1系数的指标上分别达到95.78%、83.32%和91.86%。     

Variations on Lanczos' tridiagonalization process

Lanczos' tridiagonalization processes transform a matrix into an equivalent tridiagonal one. In this paper, we propose several ways of implementing these procedures. They are based on different choices of the auxiliary polynomials which appear in the underlying theory of formal orthogonal polynomials and on a change in their normalization. We also give transpose-free variants of Lanczos processes in the spirit of the CGS and BiCGStab algorithms for solving a system of linear equations. Numerical experiments show that some of the variants proposed have a better numerical behavior than the original algorithms. Received: August 1999 / Revised version: December 1999     

多级信息补偿的U型网络图像超分辨率重建算法

针对基于深度神经网络的图像超分辨率重建算法在特征提取过程中容易丢失特征信息,导致重建图像缺少纹理和边缘细节等问题,提出一种多级信息补偿的U型网络图像超分辨率重建算法.首先设计一个用于图像超分辨率重建的U型网络,该网络通过下通道分支对输入特征进行多层级特征提取和通道压缩,通过底层模块对压缩后的特征进行融合并提取不同通道的相关特征,通过上通道分支对压缩后的相关特征进行多层次特征提取和通道恢复;然后设计多级信息补偿模型,对U型网络的通道压缩过程中丢失的信息和通道恢复过程中难以恢复的信息进行补偿;最后在不同放大倍数下的Set5、Set14、BSD100和Urban100测试集上对所提算法和主流算法进行对比测试分析,实验结果表明所提算法相比主流算法实现了在峰值信噪比(PSNR)/结构相似度(SSIM)指标和视觉效果上的巨大提升.     

面向图像修复取证的U型特征金字塔网络

图像修复是一种常见的图像篡改手段,而基于深度学习的图像修复方法能生成更复杂的结构乃至新的对象,使得图像修复取证工作更具有挑战性。因此,提出一种端到端的面向图像修复取证的U型特征金字塔网络（FPN）。首先,通过自上而下的VGG16模块进行多尺度特征提取,并利用自下而上的特征金字塔架构对融合后的特征图进行上采样,整体流程形成U型结构;然后,结合全局和局部注意力机制凸显修复痕迹;最后,使用融合损失函数以提高修复区域的预测率。实验结果表明,所提方法在多种深度修复数据集上的平均F1分数和IoU值分别为0.791 9和0.747 2,与现有的基于扩散的数字图像修复定位（LDI）、基于图像块的深度修复取证方法（Patch-CNN）和基于高通全卷积神经网络（HP-FCN）方法相比,所提方法具有更好的泛化能力,且对JPEG压缩也具有较强的鲁棒性。     

一种用于视频对象分割的仿U形网络

在半监督的分割任务中,单镜头视频对象分割(OSVOS)方法根据第一帧的对象标记掩模进行引导,从视频画面中分离出后续帧中的前景对象。虽然取得了令人印象深刻的分割结果,但其不适用于前景对象外观变化显著或前景对象与背景外观相似的情形。针对这些问题,提出一种用于视频对象分割的仿U形网络结构。将注意力机制加入到此网络的编码器和解码器之间,以便在特征图之间建立关联来产生全局语义信息。同时,优化损失函数,进一步解决了类别间的不平衡问题,提高了模型的鲁棒性。此外,还将多尺度预测与全连接条件随机场(FC/Dense CRF)结合,提高了分割结果边缘的平滑度。在具有挑战性的DAVIS 2016数据集上进行了大量实验,此方法与其他最先进方法相比获得了具有竞争力的分割结果。     

A bulk-micromachined bistable relay with U-shaped thermal actuators

This paper reports a deep-reactive ion etching (DRIE)-through-etched laterally bistable MEMS relay for power applications, with a primary emphasis on the design and modeling of its U-shaped transient thermal actuators, and a secondary emphasis on the design and fabrication of its contact element. In this relay, a contact crossbar is carried by a curved-beam bistable mechanism , which is toggled by transient U-shaped thermal actuators with their hot beam adiabaticly heated by electrical pulses. Each U-shaped thermal actuator comprises uniform-thickness hot and cold beams with a gap between them so they bend differently. This paper develops both a basic model and a complete model for the actuator that are verified by Finite Element Analysis and serve as effective design tools. The DRIE process creates nonideal etched surfaces, which pose challenges for good relay contacts. Both contact design and process development are discussed to help alleviate this problem. The fabricated relay exhibits a minimum total on-state resistance of 60 m/spl Omega/, and a maximum current carrying capacity of 3 A. It switches with a 1 ms actuation pulse, and a maximum 5 Hz repetition rate.     

高速移动下U型槽的时变信道建模

随着国内高速铁路建设的迅速发展,在高速铁路上要求移动办公、娱乐的客户需求与日俱增,而现有的蜂窝移动通信以及针对铁路的移动通信铁路全球系统（GSM-R）均不能很好地满足客户对宽带无线通信的服务质量（QoS）需求。高铁在实际的行驶过程中,会经历各种复杂的场景,U型槽是常见的场景,然而目前尚未有充分针对高速移动下U型槽的时变信道建模的研究。针对此问题,提出一种高速移动下U型槽的时变信道建模方法。首先,采用几何随机分布理论,针对高铁典型场景U型槽建立几何分布模型,分析散射体簇的变化规律,推导视距（LOS）分布、时变角度扩展、时变多普勒扩展等参数的表达式,并给出了信道冲击响应的闭式解。其次,分析了信道的时变空时域互相关函数、时变自相关函数以及时变空域多普勒功率谱密度的表达式。最后,对所提模型进行了统计性能的仿真,验证了该模型具有时变性以及较高的相关性,体现了高铁信道的非平稳性,满足高速无线信道的特性。     

Analytic Variations on the Airy Distribution

The Airy distribution (of the ``area' type) occurs as a limit distribution of cumulative parameters in a number of combinatorial structures, like path length in trees, area below walks, displacement in parking sequences, and it is also related to basic graph and polyomino enumeration. We obtain curious explicit evaluations for certain moments of the Airy distribution, including moments of orders -1 , -3 , -5 , etc., as well as +\frac13 , -\frac53 , -\frac 11 3 , etc. and -\frac73 , -\frac 13 3 , -\frac 19 3 , etc. Our proofs are based on integral transforms of the Laplace and Mellin type and they rely essentially on ``non-probabilistic' arguments like analytic continuation. A by-product of this approach is the existence of relations between moments of the Airy distribution, the asymptotic expansion of the Airy function \Ai(z) at +∈fty , and power symmetric functions of the zeros -α _k of \Ai(z) . Received June 6, 2000; revised February 17, 2001.     

Persian Variations

The concept of variation is essential in geometric design. It is surprising that patterns very different may be variations of the same model. We define two families of pentagonal patterns with three kind of variations, and give some suggestions how to analyse these patterns and create in this style. We then search for self-similarity systems in a strict sense. Although from a systematic search, the two solutions proposed here can also generate some traditional 2-level patterns. In searching for subdivisions of the tiles into rhombuses, we found two solutions. Both can be compatible with the Binary Tiling (not with the Penrose Tiling). Then, using the concept of X-Tiles defined in a previous paper (Castera et al., http://castera.net/entrelacs/public/articles/Flying_Patterns.pdf, 2011), we find new relationship between the two families of pentagonal patterns. In the last chapter we show and comment some examples taken from traditional architecture in Iran, and infer a self-similar system for pattern with interlaces from a 2-level tiling in Isfahan. This paper reflect the point of view of a pattern designer.     

Optimal allocation of heterogeneous workers in a U-shaped production line

In this paper, we deal with a U-shaped production line with multiple heterogeneous multi-function workers. Skills of workers are assumed to be different. We consider an optimization problem for finding an allocation of workers to the line that minimizes the overall cycle time under the minimum number of workers satisfying the demand. All of processing, operation and walking times are deterministic. We propose an algorithm for computing an optimal allocation of workers to machines.     

融合学习心理学的人类学习优化算法

针对简单人类学习优化（SHLO）算法寻优精度低和收敛慢的问题,提出了一种融合学习心理学的人类学习优化算法（LPHLO）。首先,结合学习心理学中的小组学习（TBL）理论引入TBL算子,从而在个体经验、社会经验的基础上,增加了小组经验来对个体学习状态进行控制,避免算法早熟收敛;然后,结合记忆编码理论提出了动态调参策略,从而实现个体信息、社会信息、团队信息的有效融合,更好地平衡了算法局部探索和全局开发的能力。选取典型的组合优化难题——背包问题中的两种算例,即单约束背包问题、多约束背包问题进行仿真实验,实验结果表明,所提LPHLO与基本的SHLO算法、遗传算法（GA）和二进制粒子群优化（BPSO）算法等算法相比,在寻优精度和收敛速度方面更具优势,具有更好的解决实际问题的能力。     

Variations on the Theme of Scarf's Counter-Example

We study the relation between the stability of a competitive equilibrium (CE)and the price adjustment mechanism used to attain that equilibrium point.Using two specific examples, a three-commodity exchange economy with a uniquecompetitive equilibrium (Scarf's global instability example) and atwo-commodity, two-trader type exchange economy with multiple competitiveequilibria, we show that the stability of a CE depends critically upon thedynamics of the price adjustment mechanism. A particular CE may be unstableunder one price adjustment mechanism but stable under another. The jointdynamics of the chosen price adjustment mechanism and the given economydetermines the overall stability of its competitive equilibrium. Our resultssuggest that context-rich studies of economic systems which focus on aspecific price adjustment mechanism may provide insights into the dynamics andstability of economic systems that are often not revealed through acontext-independent analysis.     

Stochastic analysis of a U-shaped production line with multiple workers

In a conveyor system for mass production as in the Ford system, each station processes just one item in one cycle time, where the cycle time is the time-interval between two successive outputs. In a Just-In-Time production system, a single-unit production and conveyance system is applied to a part production line without conveyors, and multi-function workers and a U-shaped layout of machines have been introduced. The multi-function worker takes charge of multiple machines. In this paper, we consider a U-shaped line with multiple multi-function workers, each of which is responsible for several machines. We formulate it into a marked Petri net, which has a convex property of a generalized semi-Markov process. This shows that the cycle times of two different systems are comparable in the sense of an increasing convex order. We also show the reversibility of the system, that is, the expected cycle time of the reversed system where each worker operates items and walks in the reversed order of stations is the same as that of the original system.