Timing of cueing in complex problem-solving tasks: Learner versus system control

Task-specific cueing formats that promote the automation and construction of problem-solving schemas should ideally be presented just in time to students learning to solve complex problems. This article reports experimental work comparing learner-controlled cueing, system-controlled cueing, and no cueing among 34 sophomore law students in a multimedia practical aimed at learning to prepare and hold a plea in court. The cueing consisted of a combination of process worksheets (PW) and worked out examples (WOE). Our main hypotheses that participants with cueing would outperform those without cueing and that participants with learner-controlled cueing would outperform those with system-controlled cueing were partly confirmed by the learning and transfer outcomes on a training and transfer task. Theoretical and practical implications of these findings are discussed.     

目标不同视角下观察信息的迁移和再利用

基于视觉信息的目标检测和识别模型在训练时往往依赖于来自于训练样本的视角信息,然而附带了视角信息的训练样本通常只有很少的数据库可以提供。当此类信息缺失时,传统的通用目标检测系统通常通过一些非监督学习方法来对样本的视角信息进行粗略估计。本文改进并引入了一种选择性迁移学习方法即TransferBoost方法来解决目标视角信息缺失的问题。本文TransferBoost方法基于GentleBoost框架实现,该方法通过重新利用其它类别样本中的先验信息来提升当前类别样本的学习质量。当给定一个标定完善的样本集作为源数据库时,TransferBoost通过同时调整每个样本的权值和每个源任务的权值实现样本级和任务级的两级知识迁移。这种双层迁移学习更有效地从混合了相关源数据和不相关源数据的数据集中提取了有用的信息。实验结果表明,和直接使用传统的机器学习方法相比较,迁移学习方法所需要的训练样本数大大减少,从而降低了目标检测与识别系统的训练代价,扩展了现有系统的应用范围。     

The impact of knowledge representation on cognitive‐oriented task performance

This research examines the impact of training style and operator individual differences on the task representation developed, automatized task performance, and controlled task performance. Results indicate that performance on relatively straightforward repetitive tasks usually associated with automatization is influenced by training style and the mental task representation held by operators. Also, domain representation is a significant determinant of performance on complex cognitive‐oriented tasks requiring controlled processes. Therefore, the task representation is identified as a high‐level performance determinant for both simple and complex task performance. No effect for training style or individual differences was found. It is concluded that training programs for systems requiring human‐computer interaction must account for this factor in order to facilitate the learning process and enhance task performance.     

Radial Basis Function Network for Multitask Pattern Recognition

In this paper, a novel type of radial basis function network is proposed for multitask pattern recognition. We assume that recognition tasks are switched sequentially without notice to a learner and they have relatedness to some extent. We further assume that training data are given to learn one by one and they are discarded after learning. To learn a recognition system incrementally in such a multitask environment, we propose Resource Allocating Network for Multi-Task Pattern Recognition (RAN-MTPR). There are five distinguished functions in RAN-MTPR: one-pass incremental learning, task change detection, task categorization, knowledge consolidation, and knowledge transfer. The first three functions enable RAN-MTPR not only to acquire and accumulate knowledge of tasks stably but also to allocate classes to appropriate tasks unless task labels are not explicitly given. The fourth function enables RAN-MTPR to recover the failure in task categorization by minimizing the conflict in class allocation to tasks. The fifth function, knowledge transfer from one task to another, is realized by sharing the internal representation of a hidden layer with different tasks and by transferring class information of the most related task to a new task. The experimental results show that the recognition performance of RAN-MTPR is enhanced by introducing the two types of knowledge transfer and the consolidation works well to reduce the failure in task change detection and task categorization if the RBF width is properly set.     

结合集成学习与迁移学习的标签比例学习方法

标签比例学习（LLP）是一种将实例放入包中的机器学习方法,它只提供包中的实例信息和标签比例信息,而不提供标签信息。针对多个相关任务的LLP问题,提出了一种基于迁移学习的标签比例集成学习模型,简称AT-LLP,该模型通过在任务之间构建共享参数来连接相关任务,将源任务中学习到的知识迁移到目标任务中,从而提高目标任务的学习效率。同时该算法引入了集成学习算法,在分类器多轮迭代的学习过程中,不断调整训练集的权重系数,进一步将弱分类器训练为强分类器。实验表明,所提AT-LLP模型比现有LLP方法具有更好的性能。     

元迁移学习在少样本跨域图像分类中的研究

目的 现有基于元学习的主流少样本学习方法假设训练任务和测试任务服从相同或相似的分布,然而在分布差异较大的跨域任务上,这些方法面临泛化能力弱、分类精度差等挑战。同时,基于迁移学习的少样本学习方法没有考虑到训练和测试阶段样本类别不一致的情况,在训练阶段未能留下足够的特征嵌入空间。为了提升模型在有限标注样本困境下的跨域图像分类能力,提出简洁的元迁移学习（compressed meta transfer learning,CMTL）方法。方法 基于元学习,对目标域中的支持集使用数据增强策略,构建新的辅助任务微调元训练参数,促使分类模型更加适用于域差异较大的目标任务。基于迁移学习,使用自压缩损失函数训练分类模型,以压缩源域中基类数据所占据的特征嵌入空间,微调阶段引导与源域分布差异较大的新类数据有更合适的特征表示。最后,将以上两种策略的分类预测融合视为最终的分类结果。结果 使用mini-ImageNet作为源域数据集进行训练,分别在EuroSAT(EuropeanSatellite)、ISIC(InternationalSkinImagingCollaboration)、CropDiseas(Cr...     

Are high‐coherent concept maps better for prior knowledge activation? Differential effects of concept mapping tasks on high school vs. university students

Abstract We investigated whether and how prior knowledge activation improves learning outcomes for high school (less experienced learners) and university students (experienced learners) in a hypertext environment. Map coherence was defined as the extent to which relationships between the concepts in the map were made explicit. Therefore, we classified the mapping task of creating and labelling lines as low‐coherent, and the mapping task of labelling provided lines as high‐coherent. Learners were randomly assigned to the conditions of (1) high‐coherent knowledge activation; (2) low‐coherent knowledge activation; and (3) a baseline condition without prior knowledge activation. We found an overall effect for prior knowledge activation, learning experience, and an interaction between learning experience and the coherence of the prior knowledge activation task on learning outcomes. High school students benefited most from labelling provided lines, while university physics majors benefited most from creating and labelling lines. This interaction effect and effects of the specific mapping tasks on process measures support the claim that different prior knowledge activation tasks are suited for different groups of learners.     

Learning and transfer effects in simulated industrial information processing tasks

Practice and alternation among a set of jobs are the characteristics of any jobbing industry. Learning and transfer of learning, which are the main human factors issues in practice and alternation, were investigated through laboratory simulation of two typical industrial information processing tasks. In the first experiment a location task was examined while a search task was investigated in experiment two. In both experiments two levels of task complexity, two groups of subjects, and two positions were combined in a 2 × 2 latin square formation with 800 trials in each task level. The results show that learning pattern appears to be task dependent, with quicker learning in location tasks than in search tasks. Learning also transfers differently for the tasks considered. Implications for the industrial training program for assembly tasks are discussed.     

Development and Evaluation of a Virtual Reality Training System Based on Cognitive Task Analysis: The Case of CNC Tool Length Offsetting

This article reports on the development and evaluation of a virtual reality training system (VRTS) for a specific machining task. A cognitive task analysis of expert machinists was conducted to examine whether this can be effective in developing a VRTS concerning tool length offsetting for a machining center. This analysis provided the necessary information for development and calibration of such a system. Subsequently, the effectiveness of the VRTS was evaluated by conducting an experiment with 29 mechanical engineering students. The VRTS set‐up comprised a video projection of the machining center and a physical mock‐up of its interface. The system demonstrated positive training transfer for the toll length offsetting task in terms of task accomplishment and of time to complete the task. No positive transfer was observed in terms of task accuracy, probably due to perceptual biases induced by the detailed specification of the VRTS. The present work provides evidence that cognitive task analysis was effective in identifying a number of key skills pertaining to the tool length offsetting task and in implementing ways to facilitate training in such tasks in a virtual environment. This article also demonstrates that even for tasks that include subtle perceptual skills VRTS may be beneficial regardless of the level of physical fidelity, provided that the cognitive organization of a task is adequately mapped in the system.     

New goals for HCI training: How to mix old and new skills in the trainee

The development of human‐computer interaction systems and the acquisition of skills associated with such systems typically occur in the context of previous experience. What is learned in one situation may facilitate or impede learning in another situation. The aim of this article is to discuss the role of experience in human‐computer interaction. The ACT? theory of skill acquisition and transfer is extended to account for the effects of old skills on the learning of new tasks. The extended model predicts a number of changes in performance that will occur when a new task involves the combination of old and new skills, including the suggestion that the learning rate of the new task will be slower than the rate at which the old skills were originally acquired. Two experiments are reported, the results of which support most of the model's predictions. The results also suggest that the minimum performance time of a task may be increased if performance of the task involves combining old and new skills. Implications of the effects of such combinations are considered with respect to the best methods of training for human‐computer interaction systems and the development of such systems.     

基于多样真实任务生成的鲁棒小样本分类方法

随着大数据、计算机与互联网等技术的不断进步,以机器学习和深度学习为代表的人工智能技术取得了巨大成功,尤其是最近不断涌现的各种大模型,极大地加速了人工智能技术在各个领域的应用.但这些技术的成功离不开海量训练数据和充足的计算资源,大大限制了这些方法在一些数据或计算资源匮乏领域的应用.因此,如何利用少量样本进行学习,也就是小样本学习成为以人工智能技术引领新一轮产业变革中一个十分重要的研究问题.小样本学习中最常用的方法是基于元学习的方法,这类方法通过在一系列相似的训练任务上学习解决这类任务的元知识,在新的测试任务上利用元知识可以进行快速学习.虽然这类方法在小样本分类任务上取得了不错的效果,但是这类方法的一个潜在假设是训练任务和测试任务来自同一分布.这意味着训练任务需要足够多才能使模型学到的元知识泛化到不断变化的测试任务中.但是在一些真正数据匮乏的应用场景,训练任务的数量也是难以保证的.为此,提出一种基于多样真实任务生成的鲁棒小样本分类方法(DATG).该方法通过对已有少量任务进行Mixup,可以生成更多的训练任务帮助模型进行学习.通过约束生成任务的多样性和真实性,该方法可以有效提高小样本分类方...     

Evaluation of two types of online help for application software

Abstract A recent trend in application software design is to extend online help systems in order to support exploratory and self‐paced learning. Two different information formats, lists of action steps that have to be taken to achieve a goal (operative help) and explanations about how a function works (function‐oriented help), were evaluated to assess their effects on learning performance. In two experiments, adult computer novices (N = 60 and N = 20) learned to use experimental graphics software by task‐based exploration. After one half of the tasks, the type of tasks changed so that the acquired action schema could no longer be applied. Results indicated that participants who had access to function‐oriented help were coping more successfully with the schema change than participants who received operative help. The relevance to the design of software training and online help systems is discussed.     

Multitask Learning

Multitask Learning is an approach to inductive transfer that improves generalization by using the domain information contained in the training signals of related tasks as an inductive bias. It does this by learning tasks in parallel while using a shared representation; what is learned for each task can help other tasks be learned better. This paper reviews prior work on MTL, presents new evidence that MTL in backprop nets discovers task relatedness without the need of supervisory signals, and presents new results for MTL with k-nearest neighbor and kernel regression. In this paper we demonstrate multitask learning in three domains. We explain how multitask learning works, and show that there are many opportunities for multitask learning in real domains. We present an algorithm and results for multitask learning with case-based methods like k-nearest neighbor and kernel regression, and sketch an algorithm for multitask learning in decision trees. Because multitask learning works, can be applied to many different kinds of domains, and can be used with different learning algorithms, we conjecture there will be many opportunities for its use on real-world problems.     

命名实体识别的迁移学习研究综述

命名实体识别(NER)是自然语言处理的核心应用任务之一.传统和深度命名实体识别方法严重依赖于大量具有相同分布的标注训练数据,模型可移植性差.然而在实际应用中数据往往都是小数据、个性化数据,收集足够的训练数据是非常困难的.在命名实体识别中引入迁移学习,利用源域数据和模型完成目标域任务模型构建,提高目标领域的标注数据量和降...     

Can a short nap and bright light function as implicit learning and visual search enhancers?

The present study examined effects of a short nap (20 min) and/or bright light (2000 lux) on visual search and implicit learning in a contextual cueing task. Fifteen participants performed a contextual cueing task twice a day (1200–1330 h and 1430–1600 h) and scored subjective sleepiness before and after a short afternoon nap or a break period. Participants served a total of four experimental conditions (control, short nap, bright light and short nap with bright light). During the second task, bright light treatment (BLT) was applied in the two of the four conditions. Participants performed both tasks in a dimly lit environment except during the light treatment. Results showed that a short nap reduced subjective sleepiness and improved visual search time, but it did not affect implicit learning. Bright light reduced subjective sleepiness. A short nap in the afternoon could be a countermeasure against sleepiness and an enhancer for visual search.

Practitioner Summary: The study examined effects of a short afternoon nap (20 min) and/or bright light (2000 lux) on visual search and implicit learning. A short nap is a powerful countermeasure against sleepiness compared to bright light exposure in the afternoon.     

Investigating the Impact of Task Change Type and Transparency on Transfer of Learning

This article investigates users' transfer of learning on two e-mail platforms with different transparency levels and task change types. The objective is to provide implication on task design in product upgrading. A mixed design of 4 (types of task change)?×?2 (transparency)?×?2 blocks (Hotmail and Outlook platform) within-subjects design and 3 levels of expertise between-subjects design, both having repeated measures on type of task change and transparency, was used to evaluate participants' transfer of learning. A main effect of both transparency and task change type is expected. High-transparency tasks are expected to lead to higher performance than low-transparency tasks. Commission task changes are expected to lead to lower performance than other types of task change due to the addition of new task steps. The results showed that commission task changes led to the greatest disturbance on user performance, whereas performance is best during sequence task changes in which participants have a positive transfer in performance time per step and no increase in error ratio. Participants also showed better transfer of learning in high-transparency tasks. The conclusion is that transparency level is critical in the future design of product changes. A refined design approach is necessary to ensure that the design tasks are highly transparent to users. In addition, to facilitate user's transfer of learning, commission task changes should be reduced and sequence task changes should be increased.     

The effects of grouping and cueing in the performance of location tasks

In location tasks such as assembly of a control panel, operators respond to stimulus information by locating a given target in an extensive set of response alternatives. Arrangement of the response alternatives and the method of presenting the location information (cueing), as well as the interaction between these factors, were hypothesised to influence performance in this type of task. To test these hypotheses, a factorial experiment involving 60 subjects was performed in which five levels of grouping and four levels of cueing were investigated. Grouping appeared to affect location accuracy more than location time, whereas the effects of cueing were found to be significant for both location accuracy and time. The absence of an interaction effect between grouping and cueing suggested that the effects of these factors on performance were independent.     

基于感知学习的成人弱视在线训练系统的设计

本文根据感知学习在成人弱视治疗中的最新研究成果,采用Gabor作为训练刺激视标,选择视标在有噪环境下的识别(Discrimination)任务作为训练方法,分析并设计了基于感知学习的成人弱视训练系统。系统给出了两种不同条件下的刺激任务对成人弱视患者进行感知学习训练,根据训练的结果调节下一个模块训练的难度系数,降低患者对比度阈值(Contrast Threshold),提高视觉对比敏感度(Contrast Sensitivity)。考虑到患者长期在医院治疗带来的不便,系统提出了基于网络支持的体系结构,将训练数据同步到中心数据库,为计算下一个模块的训练参数提供保证,并使患者可以随时在家中进行训练治疗。     

双Q网络学习的迁移强化学习算法

深度强化学习在训练过程中会探索大量环境样本,造成算法收敛时间过长,而重用或传输来自先前任务(源任务)学习的知识,对算法在新任务(目标任务)的学习具有提高算法收敛速度的潜力.为了提高算法学习效率,提出一种双Q网络学习的迁移强化学习算法,其基于actor-critic框架迁移源任务最优值函数的知识,使目标任务中值函数网络对策略作出更准确的评价,引导策略快速向最优策略方向更新.将该算法用于Open AI Gym以及在三维空间机械臂到达目标物位置的实验中,相比于常规深度强化学习算法取得了更好的效果,实验证明提出的双Q网络学习的迁移强化学习算法具有较快的收敛速度,并且在训练过程中算法探索更加稳定.     

An informational perspective on skill transfer in human-machine systems

Differentiation of perceptual invariants is proposed as a theoretical approach to explain skill transfer for control at the human-machine interface. I propose that sensitivity to perceptual invariants is enhanced during learning and that this sensitivity forms the basis for transfer of skill from one task to another. The hypothesis implies that detection and discrimination of critical features, patterns, and dimension of difference are important for learning and for transfer. This account goes beyond other similarity conceptions of transfer. To the extent that those conceptions are specific, they cannot account for effects in which performance is better following training on tasks that are less rather than more similar to the criterion task. In essence, this is a theory about the central role of low-dimensional informational patterns for control of behavior within a high-dimensional environment, and about the adjustment of an actor's sensitivity to changes in those low-dimensional patterns.