元认知作为认知的核心，包括客体水平和元水平，两种水平之间的信息加工的形式将元认知分为监测和控制两种成分。较高水平的元认知能力，意味着对行为的有效监测与控制。考虑到提升元认知能力带来的广泛好处，本研究基于强化学习理论，采用即时反馈调节被试自信度主观报告的训练模式，并设计了若干基于感知觉刺激的任务，以期探讨个体的元认知能力是否存在训练相关的变化。在本研究中共有三个实验：
实验一采用基于随机点运动的元认知训练任务，其目的是为了验证了基于强化学习的元认知训练模式的效果。实验二中增加了随机点运动、随机点颜色的决策-再决策(decision-redecision)任务作为训练的前后测，其目的是验证在相同的刺激下，元认知能力的不同成分间是否存在相关性，即训练元认知的监测能力能否即时提高被试的元认知控制能力。除此之外，实验二还旨在探讨元认知能力的相同成分在不同知觉刺激之间是否存在一致性，即在特定任务中训练元认知的监测能力能否即时提高被试在其他任务中的元认知监测能力。实验三采用短期的基于随机点运动的元认知训练任务，其目的为探讨基于强化学习的元认知能力训练过程中影响训练效果稳定性的因素。
实验一的结果表明，基于强化学习的元认知训练模式，可以帮助被试的元认知能力在一定条件下随训练时间的增加而得到一定程度上的提高。利用包含多个水平感知觉刺激难度的训练方法可以优化被试的训练效果。这是由于当刺激难度不再单一时，被试可以通过刺激难度差异进行相互参考，从而在整体水平上报告出与决策层面更加一致的自信度。通过实验二的结果，研究发现，在基于强化学习的元认知能力训练的前后，跨知觉刺激的前后测任务的结果表明，被试整体上的元认知调控能力并没有表现出训练前后的显著差异，即训练元认知的监测能力不能即时提高被试的元认知控制能力。同时在跨知觉刺激的前后测任务中，被试整体上的元认知监测能力表现出了训练前后的显著差 异，即不同刺激形式的任务前后测中 decision 阶段 AUC 的差异存在区别，这证明被试在某种特定任务下被训练出的元认知监测能力不能很好的迁移到其他任务中。实验三的结果表明，元认知能力训练效果与个体元认知敏感性的差异存在负相关，元认知敏感性较高的被试元认知能力训练效果更差。最后，综合三个实验的结果，发现基于强化学习的元认知能力训练过程中的训练效果并不显著，强化学习对于元水平能力的调整存在一定的挑战。

As the core of cognition, metacognition includes object-level and meta-level. The form of information processing between these two levels divides metacognition into monitor and control. A higher level of metacognitive ability means the effective monitoring and control of behavior. Considering the broad benefits of improving metacognitive ability, this study is based on the theory of reinforcement learning. To explore whether the individual’s metacognitive ability is related to changes in cognitive abilities, the study uses instant feedback to adjust the training mode of participants’ self-confidence reports and constructs several tasks based on perceptual stimuli. There are three experiments in this study:
The first experiment is a metacognitive training experiment based on perception stimuli - random dots motion. The purpose of the first experiment is to verify the effect of the metacognitive training model based on reinforcement learning. The second experiment is the decision-re-decision task based on random dots motion stimuli and random dots color stimuli. The purpose of the second task is mainly to verify whether there is a correlation between the different components of metacognitive ability under the same stimulus, that is, whether training the metacognitive monitor ability of the subjects can instantly improve the metacognitive control ability of the subjects. The purpose of the second experiment is also mainly to verify whether there is consistency among the same components of metacognitive ability in different perceptual stimuli, that is, to exam whether training of metacognitive monitor ability in a particular task, can immediately improve the metacognitive monitor ability in other tasks. The third experiment mainly uses short-term metacognitive training tasks based on random point exercises, and its purpose is to explore the factors that affect the stability of training effects in the process of metacognitive ability training.
The results of the first experiment in the experiment one show that the metacognitive training model based on reinforcement learning can help the subjects' metacognitive ability to be improved to a certain extent with the increase of training time under certain conditions.
Moreover, the use of multiple levels of difficulty training methods can optimize the subjects' training effects. When the difficulty of the stimuli is no longer single, the subjects can refer to each other through the difficulty of the stimuli, and thereby report a confidence level that is more consistent with the decision-making level at the overall level. The results of the second experiment finds that the difference between the perceptual stimuli will affect the participants' metacognitive evaluation, rather than the accuracy of cognitive decision-making. The results of the cross-perceptual stimulus task indicate that there is no significant difference between the metacognitive control ability of the subjects before and after training, Hence, the training of metacognitive monitor ability could not immediately improve the metacognitive control ability of the subjects. At the same time, in the cross-perceptual stimulus task, the subjects’ metacognitive monitoring ability showed significant differences before and after the training, that is, the difference in the AUC of the decision stage before and after the training in the cross- perceptual stimulus task, which proves that the metacognitive monitor ability trained under this specific task cannot be transferred to other tasks well. The results of the third experiment show that the effect of metacognitive ability training is negatively correlated with the difference of individual metacognitive sensitivity, and the effect of metacognitive ability training is worse for subjects with higher metacognitive sensitivity. Finally, combining the results of the three experiments, it is found that the training effect in the process of metacognitive ability training based on reinforcement learning is not significant, and reinforcement learning poses certain challenges to the adjustment of meta-level ability.