深度学习的发生机制与多模态数据测评研究

Research on the Concurrence Mechanism and Multi-modal Evaluation of Deep Learning

进入数字化知识经济时代,社会发展对教育活动提出了更高要求,旨在培养高阶思维和创新能力的深度学习备受国际社会的高度关注。为此,“深度学习如何发生”和“如何评估深度学习发生程度”,也成为亟待研究的两个关键学术问题。认知心理学、教育神经学和具身认知为理解深度学习提供了理论基础,指引人们从人—物互动视角认识个体的认知变化过程,打开了个体学习的内隐机制“黑箱”,实现内隐机制和外显表征的桥接与理解。基于这一认识与分析,通过信息输入、深度加工和学习生成三个环节,初步构建了深度学习的发生机制模型,并设置相关的学习状态指标。而要实现对深度学习发生程度的精准评价,则应依据深度学习的多模态数据测评框架,综合性采集与分析学习者的生理数据、自我评估数据、在线学习平台数据和课堂参与数据等多模态数据。但是,在采用多模态数据测评深度学习的发生及其程度时,多模态数据仍存在异质性差距、数据建模缺乏精确度、常态化评价开展困难、数据安全和伦理道德难以得到保障等问题,亟需未来开展更为深入的研究与探索。

In the era of knowledge economy characterized with digitization,the society has put forward higher requirements for educational activities.The international community has paid attention to deep learning which concentrated on cultivating students’high-order thinking and innovative ability.Moreover,the two key issues of“how deep learning occurs”and“how to evaluate the degree of deep learning”have always been urgent to be solved.Cognitive psychology,educational neuroscience and embodied cognitive theory provide the basis for understanding deep learning.These theories guide us to understand the individual cognitive change process from the perspective of human-object interaction,open the"black box"of the implicit mechanism of individual learning,and realize the bridging and understanding of implicit mechanism and explicit representation.Based on this understanding and analysis,through three links of information input,deep processing and learning generation,the study preliminarily construct the occurrence mechanism model of deep learning and set relevant learning state indicators.Additionally,in order to accurately evaluate the degree of deep learning,the study comprehensively set the the multi-modal data evaluation framework of deep learning.The relevant multi-modal data of the framework is consist of the learners’physiological data,self-assessment data,online learning platform data and class participation data etc.However,when using multi-modal data to evaluate the occurrence and degree of deep learning,there are still some problems,such as the heterogeneity gaps in multi-modal data,the lack of accuracy in data modeling,the difficulties in carrying out normalized evaluation and in ensuring data security and ethics.These problems urgently need more in-depth research and exploration in the future.