Attentional retraining procedures: Manipulating early or late components of attentional bias?

According to cognitive models of anxiety disorders, attentional bias for threatening information is a vulnerability factor to the etiology and maintenance of anxiety. A recently developed methodology to reduce attentional bias has been found to reduce emotional reactivity and anxiety. The present study aimed at identifying the effects of this attentional bias reduction on early and later stages of threat processing. Undergraduates were allocated to an attentional bias reduction (n = 23) versus control condition (n = 25). It was found that attentional bias reduction influenced late but not early stages of threat processing. This finding is of theoretical importance in relation to studies on the causal role of attentional bias and emotional reactivity. Moreover, the present findings also bear relevance to the clinical application of attentional retraining procedures. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

A question of belonging: Race, social fit, and achievement.

Stigmatization can give rise to belonging uncertainty. In this state, people are sensitive to information diagnostic of the quality of their social connections. Two experiments tested how belonging uncertainty undermines the motivation and achievement of people whose group is negatively characterized in academic settings. In Experiment 1, students were led to believe that they might have few friends in an intellectual domain. Whereas White students were unaffected, Black students (stigmatized in academics) displayed a drop in their sense of belonging and potential. In Experiment 2, an intervention that mitigated doubts about social belonging in college raised the academic achievement (e.g., college grades) of Black students but not of White students. Implications for theories of achievement motivation and intervention are discussed. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

动作识别中基于深度神经网络和GA合并算法的分类决策方法

针对人体动作识别中传统方法在分类决策方面存在问题和缺陷,提出了一种新颖的基于深度神经网络(DNN)和遗传算法(GA)合并算法的非线性分类决策方法。首先,提出的合并算法在整个训练集合上对特征提取器进行组合,进而组合成不同的两个独立网络;再利用DNN对两个独立网络进行初始化,进一步利用GA对两个网络进行合并。然后将网络的偏差和权重表示为每层网络间的一个矩阵;最后,利用DNN对网络的偏差和权重进行训练,并在合并过程中将矩阵中的每一行当作一个染色体。实验采用了标准MNIST数据集对提出算法的性能进行评估。评估结果显示实验过程中的交叉和突变操作增加了神经元节点,提高了识别性能,并且弱化了不相关和相关神经元节点。因此,提出算法的错误率更低,网络性能更优异。     

Causal attributions in posttraumatic stress disorder: Implications for clinical research and practice.

In this paper we seek to illuminate connections among basic research findings in cognition and causal inference, clinical research on the treatment of Posttraumatic Stress Disorder (PTSD), and the practices of clinicians who work with trauma survivors. We examine one particular (and, we believe, important) aspect of PTSD: The creation and maintenance of causal attributions about trauma. We suggest that elements of two principal theories of causal induction (the connectionist model and the "Power PC" causal power model) clarify the role of causal attributions in creating and sustaining the symptoms of PTSD. By exploring the role of causal attributions in creating and sustaining posttraumatic symptoms, we hope to understand better the subjective experience of trauma and its sequelae. We then suggest new directions for clinical research on cognitive restructuring in PTSD patients as well as ideas for optimizing attribution-based therapies for trauma survivors. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

基于改进的ISODATA的超球覆盖仿生模式分类算法

现有仿生模式识别分类器难以解决含有多个聚集点、非线性和稀疏性样本的分类问题。因此,引入特征分类贡献度,提出了基于改进的迭代自组织数据分析(M-ISODATA)的超球覆盖仿生模式识别算法。首先引入马氏距离对自组织数据分析方法(ISODATA)的欧氏距离替换,并引入熵权法对马氏距离进行加权以赋予各特征不同的贡献度;同时为了去除干扰样本点,引入改进的局部离群因子检测方法(M-LOF)对样本进行训练,减少了不同类别流形之间的重叠区域。再利用改进的自组织数据分析方法(M-ISODATA)对每类训练样本点动态聚类,寻找到同一类的多个小类覆盖区中心后,用超球进行该类的有效覆盖,并对落入重叠区域的测试样本点进行二次划分,实现测试样本的正确分类。最后在iris数据集上验证该算法的有效性,并将该算法应用于雷达辐射源信号的分类识别。实验结果表明,该算法具有很好的拒识、免重训能力,对于雷达信号的识别率能达到97.29%,相比于传统典型模式识别算法具有更好的识别能力。     

深度神经网络修复策略综述

随着智能信息时代的发展,深度神经网络在人类社会众多领域中的应用,尤其是在自动驾驶、军事国防等安全攸关系统中的部署,引起了学术界和工业界对神经网络模型可能表现出的错误行为的担忧.虽然神经网络验证和神经网络测试可以提供关于错误行为的定性或者定量结论,但这种事后分析并不能防止错误行为的发生,如何修复表现出错误行为的预训练神经网络模型依然是极具挑战性的问题.为此,深度神经网络修复这一领域应运而生,旨在消除有缺陷的神经网络产生的错误预测,使得神经网络满足特定的规约性质.目前为止,典型的神经网络修复范式有3种:重训练、无错误定位的微调和包含错误定位的微调.介绍深度神经网络的发展和神经网络修复的必要性;厘清相近概念;明确神经网络修复的挑战;详尽地调研目前已有的神经网络修复策略,并对内在联系与区别进行分析和比较;调研整理神经网络修复策略常用的评价指标和基准测试;展望未来神经网络修复领域研究中需要重点关注的可行方向.     

Bias for the (un)attractive self: On the role of attention in causing body (dis)satisfaction.

Objective: Body dissatisfaction plays a key role in the maintenance of eating disorders, and selective attention might be crucial for the origin of body dissatisfaction. A. Jansen, C. Nederkoorn, and S. Mulkens (2005) showed that eating disorder patients attend relatively more to their own unattractive body parts, whereas healthy controls attend relatively more to their own attractive body parts. In 2 studies, we investigated whether this bias in selective attention is causal to body dissatisfaction and whether an experimentally induced bias for attractive body parts might lead to increased body satisfaction in women who are highly dissatisfied with their bodies. Design: We used a between-subjects design in which participants were trained to attend to either their self-defined unattractive body parts or their self-defined attractive body parts by use of an eye tracker. Main Outcome Measures: State body and weight satisfaction. Results: Inducing a temporary attentional bias for self-defined unattractive body parts led to a significant decrease in body satisfaction and teaching body-dissatisfied women to attend to their own attractive body parts led to a significant increase in body satisfaction. Conclusion: Selective attention for unattractive body parts can play a role in the development of body dissatisfaction, and changing the way one looks may be a new way for improving body dissatisfaction in women. (PsycINFO Database Record (c) 2011 APA, all rights reserved)     

Learning how to grasp based on neural network retraining

Humans have an incredible capacity to manipulate objects using dextrous hands. A large number of studies indicate that robot learning by demonstration is a promising strategy to improve robotic manipulation and grasping performance. Concerning this subject we can ask: How does a robot learn how to grasp? This work presents a method that allows a robot to learn new grasps. The method is based on neural network retraining. With this approach we aim to enable a robot to learn new grasps through a supervisor. The proposed method can be applied for 2D and 3D cases. Extensive object databases were generated to evaluate the method performance in both 2D and 3D cases. A total of 8100 abstract shapes were generated for 2D cases and 11700 abstract shapes for 3D cases. Simulation results with a computational supervisor show that a robotic system can learn new grasps and improve its performance through the proposed HRH (Hopfield-RBF-Hopfield) grasp learning approach.     

Investigating the prediction capabilities of the simple recurrent neural network on real temporal sequences

The goal of this paper is to evaluate the prediction capabilities of the simple recurrent neural network (SRNN). The main focus is on the prediction of non-orthogonal vector components of real temporal sequences. A prediction problem is formulated in which the input is a component of a real sequence and the output is a prediction of the next component of the sequence. A method is developed to train a single SRNN to predict the components of sequences belonging to multiple classes. The selection of a distinguishing initial context vector for each class is proposed to improve the prediction performance of the SRNN. A systematic method to re-train the SRNN with noisy exemplars is developed to improve the prediction generalization of the network. Through the methods developed in the paper, it is demonstrated that: (a) a single SRNN can be trained to predict, contextually, the components of real temporal sequences belonging to different classes, (b) the prediction error of the SRNN can be decreased by using a distinguishing initial context vector for each class, and (c) the prediction generalization of the SRNN can be increased significantly by re-training the network with noisy exemplars.     

More rapid associative change with retraining than with initial training.

Retraining was compared with initial acquisition in 4 magazine approach experiments with rats and 1 autoshaping experiment with pigeons. The levels of performance were matched prior to reinforcement and nonreinforcement for stimuli with a history of both training and extinction and stimuli with only a minimal history of training. Under these conditions, a previously extinguished stimulus was more vulnerable both to the incremental effects of reinforcement and to the decremental effects of nonreinforcement compared with a stimulus that had only been minimally trained. (PsycINFO Database Record (c) 2010 APA, all rights reserved)