汉字识别的计算机模拟

本文总结了作者近年来在连结主义的理论框架下 ,用计算机模拟汉字识别的工作。文章分三部分。第一部分说明了计算机模拟与人工智能的关系。第二部分介绍了作者提出的两个模型 :汉字识别与命名的连结主义模型和基于语义的词汇判断的计算模型。两个模型分别成功地模拟了汉字识别中的频率效应、形声字读音中的规则效应、声旁效应、语义启动效应、语境与频率的交互作用等。第三部分讨论了模拟工作的意义、分布表征、学习算法等问题。研究表明 :认知的计算机模拟能验证人类认知实验的结果 ,对结果提出合理的解释 ,并能指导进一步的实验研究。     

异步异构分层并行演化算法及其在串并系统可靠性优化中的应用(英文)

本文提出了一种新的并行演化算法――异步异构分层并行演化算法(AHHPGA),用于求解串并系统可靠性分配的多目标优化问题(RAP)。AHHPGA采用分层结构,上层由粗粒度模型构成,下层由细粒度模型构成。在AHHPGA的每个子种群中,采用不同的全局/局部搜索度和拓扑结构,构成异构模型。AHHPGA的迁移方式包括异步接收和异步迁出。采用基于模型的复杂系统可靠性评价工具(HiP-HOPS),以克服可靠性框图方法(RBD)的缺点。仿真结果表明,基于异步异构分层并行演化算法和HiP-HOPS工具的串并系统可靠性分配的多目标优化方法,优于传统的基于遗传算法的优化方法。     

对变化/分割模型的检验(I)

用两个实验对变化／分割模型进行了检验。实验１控制目标时距和该时距内的填充数字系列,操纵数字系列的分割段数;实验２控制目标时距内的填充数字系列的间距,操纵目标时距及其中填充数字系列的分割段数,要求被试用再现法和多数估计法分别复制目标时距,并进行立即估计和延迟估计。结果显示：与存储容量模型和加工时间模型相比较,变化／分割模型对时间估计的解释具有更高的预测效度     

一种特殊的计算机指点装置追踪球的操作活动研究

探讨在指点定位中目标方向、大小和距离对食指和拇指操纵追踪球的运动时间的效应 ,并由此建立了两种追踪球的菲茨模型。用绩效指数比较了不同指点装置的操作绩效 ,对追踪球的优化设计提出了建议。     

注意促进效应依赖于项目-背景的质心间距

近期研究者发现, 在某些条件下增加对第一任务的注意能够提高伴随发生的第二任务的成绩, 并将其命名为注意促进效应。研究者提出双任务交互模型对之进行解释, 认为该效应产生的必要条件是时间上的重叠, 而不受空间距离的影响。基于对这一观点的质疑, 本研究探究了检测项目与背景图片的质心间距对注意促进效应的影响。结果发现, 当目标项位于背景图片中央时(质心间距为0), 表现出注意促进效应; 而当二者分开一定的空间距离时, 出现了注意干扰/抑制效应。由此可见, 项目和背景的质心间距会对注意促进效应产生影响。     

鼻腔流场数值模拟与鼻声反射相关性研究

探讨应用鼻声反射测量评估基于CT图像的三维重建人体鼻腔模型的可靠性,并用有限元方法对模型的流场进行数值模拟.基于志愿者CT医学图像,用表面重建的方法对人体鼻腔进行三维重建,并用有限元方法对腔体中气体的流动进行数值模拟及分析;采用鼻声反射仪测试得到志愿者鼻声反射曲线,分别提取并对比两种方法得到的鼻腔多个轴向横截面面积数据.结果显示,两种方法得到的数据有很好的吻合.由此得出结论,所建模型较真实的反映了鼻腔实际解剖结构形态,数值模拟结果可靠,对临床上呼吸道解剖结构与功能相关疾病的诊断、发病机理的深入探讨具有参考价值.     

鼻腔流场数值模拟与鼻声反射相关性研究

探讨应用鼻声反射测量评估基于CT图像的三维重建人体鼻腔模型的可靠性,并用有限元方法对模型的流场进行数值模拟。基于志愿者CT医学图像,用表面重建的方法对人体鼻腔进行三维重建,并用有限元方法对腔体中气体的流动进行数值模拟及分析;采用鼻声反射仪测试得到志愿者鼻声反射曲线,分别提取并对比两种方法得到的鼻腔多个轴向横截面面积数据。结果显示。两种方法得到的数据有很好的吻合。由此得出结论,所建模型较真实的反映了鼻腔实际解剖结构形态。数值模拟结果可靠,对临床上呼吸道解剖结构与功能相关疾病的诊断、发病机理的深入探讨具有参考价值。     

成就动机的多重目标理论

基于成就目标的三分法,Harackiewicz等人提出多重目标理论。该理论认为,学生在学习过程中采用多种类型的成就目标将会带来最佳的适应性结果,这种观点与单纯强调掌握目标正性作用的传统目标理论相对立。根据多重目标对成就结果的作用模式,理论学家分别提出了四种模型：累加目标模型、交互目标模型、特定目标模型以及选择目标模型。最近的实证研究支持了选择目标模型的观点,并为如何提升个体的成就水平提供了重要的理论依据。     

人声加工的神经机制

人声是人类听觉环境中最熟知和重要的声音, 传递着大量社会相关信息。与视觉人脸加工类似, 大脑对人声也有着特异性加工。研究者使用电生理、脑成像等手段找到了对人声有特异性反应的脑区, 即颞叶人声加工区(TVA), 并发现非人类动物也有类似的特异性加工区域。人声加工主要涉及言语、情绪和身份信息的加工, 分别对应于三条既相互独立又相互作用的神经通路。研究者提出了双通路模型、多阶段模型和整合模型分别对人声的言语、情绪和身份加工进行解释。未来研究需要进一步讨论人声加工的特异性能否由特定声学特征的选择性加工来解释, 并深入探究特殊人群(如自闭症和精神分裂症患者)的人声加工的神经机制。     

改进3PL模型参数估计的MCMC算法

本文首先用马尔科夫链蒙特卡洛(MCMC)算法和EM算法进行IRT模型参数估计模拟实验,并探讨了两种算法的参数估计精度,然后在分析三参数Logistic(3PL)模型参数估计精度的基础上改进模型并对其进行参数估计。结果表明,MCMC算法估计IRT模型的参数精度均优于EM算法,并且MCMC算法在估计3PL模型参数方面具有更明显的优势;在样本量较小的情况下,MCMC算法能较好地估计3PL模型参数,估计精度略低于2PL模型;3PL模型的项目参数确定性低是参数估计精度略低于2PL模型的主要原因;采用改进模型可以提高项目参数的确定性,进而得到更优的参数估计精度。     

Parallel search for conjunctions with stimuli in apparent motion

A series of experiments was conducted to determine whether apparent motion tends to follow the similarity rule (i.e. is attribute-specific) and to investigate the underlying mechanism. Stimulus duration thresholds were measured during a two-alternative forced-choice task in which observers detected either the location or the motion direction of target groups defined by the conjunction of size and orientation. Target element positions were randomly chosen within a nominally defined rectangular subregion of the display (target region). The target region was presented either statically (followed by a 250 ms duration mask) or dynamically, displaced by a small distance (18 min of arc) from frame to frame. In the motion display, the position of both target and background elements was changed randomly from frame to frame within the respective areas to abolish spatial correspondence over time. Stimulus duration thresholds were lower in the motion than in the static task, indicating that target detection in the dynamic condition does not rely on the explicit identification of target elements in each static frame. Increasing the distractor-to-target ratio was found to reduce detectability in the static, but not in the motion task. This indicates that the perceptual segregation of the target is effortless and parallel with motion but not with static displays. The pattern of results holds regardless of the task or search paradigm employed. The detectability in the motion condition can be improved by increasing the number of frames and/or by reducing the width of the target area. Furthermore, parallel search in the dynamic condition can be conducted with both short-range and long-range motion stimuli. Finally, apparent motion of conjunctions is insufficient on its own to support location decision and is disrupted by random visual noise. Overall, these findings show that (i) the mechanism underlying apparent motion is attribute-specific; (ii) the motion system mediates temporal integration of feature conjunctions before they are identified by the static system; and (iii) target detectability in these stimuli relies upon a nonattentive, cooperative, directionally selective motion mechanism that responds to high-level attributes (conjunction of size and orientation).     

Analysis of human body dynamics in simulated rear-end impacts

The objective of this study is to simulate the dynamic response of the human body within a rear-end impacted vehicle. A nonlinear mathematical model of a human body and a restraint system has been formulated. The model consists of connected rigid bodies representing the torso and limbs of the human frame. Nonlinear springs and dampers are used at the connection joints to represent human anatomical characteristics and limits imposed by muscles, ligaments, and soft tissue. Equations of motion are written for this model by using Kane's equation and multibody dynamics analysis procedures developed by Huston. The equations are integrated numerically for a number of specific cases where experimental data are available. Results show excellent agreement between the model and the experiments. The results of several accident simulations are also presented.     

Dynamic representations of human body movement

Psychophysical and neurophysiological studies suggest that human body motions can be readily recognized. Human bodies are highly articulated and can move in a nonrigid manner. As a result, we perceive highly dissimilar views of the human form in motion. How does the visual system integrate multiple views of a human body in motion so that we can perceive human movement as a continuous event? The results of a set of priming experiments suggest that motion can readily facilitate the linkage of different views of a moving human. Positive priming was found for novel views of a human body that fell within the path of human movement. However, no priming was observed for novel views outside the path of motion. Furthermore, priming was restricted to those views that satisfied the biomechanical constraints of human movement. These results suggest that visual representation of human movement may be based upon the movement limitations of the human body and may reflect a dynamic interaction of motion and object-recognition processes.     

Human and machine perception of biological motion

More than 30 years ago, Johansson was the first to show that humans are capable of recovering information about the identity and activity of animate creatures rapidly and reliably from very sparse visual inputs – the phenomenon of biological motion. He filmed human actors in a dark setting with just a few strategic points on the body marked by lights – so-called moving light displays (MLDs). Subjects viewing the MLDs reported a vivid impression of moving human forms, and were even able to tell the activity in which the perceived humans were engaged. Subsequently, the phenomenon has been widely studied and many attempts have been made to model and to understand it. Typical questions that arise are: How precisely is the sparse low-level information integrated over space and time to produce the global percept, and how important is world knowledge (e.g., about animal form, locomotion, gravity, etc.)? In an attempt to answer such questions, we have implemented a machine-perception model of biological motion. If the computational model can replicate human data then it might offer clues as to how humans achieve the task. In particular, if it can do so with no or minimal world knowledge then this knowledge cannot be essential to the perception of biological motion. To provide human data for training and against which to assess the model, an extensive psychophysical experiment was undertaken in which 93 subjects were shown 12 categories of MLDs (e.g., normal, walking backwards, inverted, random dots, etc.) and were asked to indicate the presence or absence of natural human motion. Machine perception models were then trained on normal sequences as positive data and random sequences as negative data. Two models were used: a k-nearest neighbour (k-NN) classifier as an exemplar of ‘lazy’ learning and a back-propagation neural network as an exemplar of ‘eager’ learning. We find that the k-NN classifier is better able to model the human data but it still fails to represent aspects of knowledge about body shape (especially how relative joint positions change under rotation) that appear to be important to human judgements.     

The representation of nonuniform motion: induced movement

Induced motion occurs when there is a misallocation of nonuniform motion. Theories of induced motion are reviewed with respect to the model for uniform motion recently proposed by Swanston, Wade, and Day. Theories based on single processes operating at one of the retinocentric, orbitocentric, egocentric, or geocentric levels are not able to account for all aspects of the phenomenon. It is therefore suggested that induced motion is a consequence of combining two different types of motion signals: one provides information by registering the motion with respect to the retina, orbit, and egocentre; the other provides information only on the relational motions between the pattern elements. Simple rules are given for defining a frame of reference for the relational motion process, which can result in a reallocation of the motion signals. It is proposed that the two signals in combination are weighted differentially, with the greater influence coming from the relational signals. Procedures for determining the weighting factors are described, and predictions from the model are examined.     

Priming and Recognition of Human Motion Patterns

Left-right orientation and size incongruence is known to affect recognition memory for objects but not object priming. In the present study, the effects of study-test changes in left-right orientation and size on old-new recognition decisions and long-term priming of human motion patterns were examined. Experiment 1 showed effects of orientation incongruence on both recognition and priming. Experiment 2 showed an effectof size incongruence on recognition memory but not on priming. It is suggested that the representations of human actions that underlie human motion priming are on a level that preserve orientation, possibly because of the importance of dynamic information for perceiving motion patterns or because encoding of human motion is governed by a body schema (e.g. Reed & Farah, 1995). In contrast, low-level metric information such as size is inconsequential to priming because priming involves identification of shape, which is not affected by size transformations. The effect of size on recognition memory, on the other hand, shows thatexplicitrecognition decisions may draw on any available episodic information, including metric attributes, to make an old new discrimination.     

The relationship between space and time in the perception of stimuli moving behind a slit

When a shape defined by a set of dots plotted along its contour is presented in a sequence of frames within the boundaries of a slit, and in each frame only one dot (featureless frame) or two dots (feature frame) are displayed, a whole moving dotted shape is perceived. Masking techniques and psychophysical measures have been used to show that a dynamic random-dot mask interferes with shape identification, provided the interframe interval is greater than about 15 ms, and there are no stimulus features for recognition in individual frames. A similar pattern of results was obtained when the observer had only to detect the movement of a single dot or a pair of dots against a dynamic-noise background. It is concluded that the visual system can resolve the correspondence problem in both apparent movement (one moving dot) and aperture viewing (featureless-frame condition) by extracting motion before the extraction of features in each frame. However, the results also show that where feature identification in each frame is possible, it can also be used to identify the moving targets.     

An oscillatory correlation model of visual motion analysis

We describe and evaluate a model of motion perception based on the integration of information from two parallel pathways: a motion pathway and a luminance pathway. The motion pathway has two stages. The first stage measures and pools local motion across the input animation sequence and assigns reliability indices to these pooled measurements. The second stage groups locations on the basis of these measurements. In the luminance pathway, the input scene is segmented into regions on the basis of similarities in luminance. In a subsequent integration stage, motion and luminance segments are combined to obtain the final estimates of object motion. The neural network architecture we employ is based on LEGION (locally excitatory globally inhibitory oscillator networks), a scheme for feature binding and region labeling based on oscillatory correlation. Many aspects of the model are implemented at the neural network level, whereas others are implemented at a more abstract level. We apply this model to the computation of moving, uniformly illuminated, two-dimensional surfaces that are either opaque or transparent. Model performance replicates a number of distinctive features of human motion perception.     

Parallel computations for controlling an arm

In order to control a reaching movement of the arm and body, several different computational problems must be solved. Some parallel methods that could be implemented in networks of neuron-like processors are described. Each method solves a different part of the overall task. First, a method is described for finding the torques necessary to follow a desired trajectory. The methods is more economical and more versatile than table look-up and requires very few sequential steps. Then a way of generating an internal representation of a desired trajectory is described. This method shows the trajectory one piece at a time by applying a large set of heuristic rules to a "motion blackboard" that represents the static and dynamic parameters of the state of the body at the current point in the trajectory. The computations are simplified by expressing the positions, orientations, and motions of parts of the body in terms of a single, non-accelerating, world-based frame of reference, rather than in terms of the joint-angles or an egocentric frame based on the body itself.