| 其他摘要 | Perceptual learning refers to steady and stable performance improvement or sensitivity enhancement as a result of repetitive training. Perceptual learning is usually long-lasting but how repetitive practice leads to long-term improvement remains largely unknown. To answer how learning cumulates as training proceeds, one of the fundamental questions in perceptual learning, the current thesis proposed a multi-component framework that covered three hypotheses: (1) improved perceptual task performance through perceptual learning may reflect cumulative effects of both long- (e.g. general learning) and short- term processes (e.g.between-session gain, between-session forgetting, within-session relearning and within-session adaptation), (2) learning different tasks may engage different longand short-term processes, and (3) coarse-grain analysis may have obscured some short-term processes while fine-grained analysis of the learning curve at the block level (e.g. tens of trials) may reveal some important long- and short-term processes.
We conducted two series of studies to examined our model and apply the model to investigate learning mechanisms under different experimental settings. In Study 1, we validated the potential of understanding a set of learning curves with the proposed multi-component model. We trained 49 subjects to learn 7 tasks sequentially in 35 days (sessions) and fitted learning curves in both coarse (e.g.session) and fine (e.g. block) grains. The session-wise analysis revealed a significant learning effect across sessions. Insterestingly, block-wise analysis not only confirmed block-by-block general learning effects but also indicated within-session relearning for all the seven tasks. In addition, between-session forgetting was identified in learning the Vernier offset discrimination, face view discrimination, and auditory frequency discrimination tasks, between-session off-line gain in the visual shape search and contrast detection tasks, and within-session adaptation in the contrast detection task. These results provided compelling evidence for our hypotheses and the rationality of our multi-component model of perceptual learning.
In Study 2, we conducted 5 experiments and applied the multi-component model to investigate how training settings, including between-session sleep, the amount of training, experience, task difficulty, external feedback, and mixed training paradigm, modulated long- and/or short-term processes.
In Experiment 2, we found no difference in the magnitude of general learning, between-session forgetting, and within-session relearning between sleep and wake groups, indicating of little influence of sleep on the between-session effect in Vernier task. Experiment 3 showed that, compared to conventional daily training group, 5-trial relearning group showed greater between-session forgetting, greater within-session relearning amplitude, and slower relearning rate in auditory frequency discrimination task, indicating that relearning process also required a critical amount of training and the amplitude of forgetting increased as between-session delay prolongs. In Experiment 4, we compared learning curves in each task between subjects who learned earlier (1st, 2nd, 3rd) and those who learned later (5th, 6th, 7th) from Study 1. The results showed that training experience affected initial performance and learning rate of general learning process and the magnitude of between-session effects. For subjects who were trained later, they exhibited worse initial performances in contrast detection, Vernier offset discrimination, motion direction discrimination, auditory frequency discrimination tasks, and slower learning rate in Vernier and face view discrimination tasks; worse initial performance but faster learning rate in N-back task. The findings were confirmed by an additional control experiment that involved only shape and Vernier tasks. Our results indicated that training experience mainly imposed a negative influence on primary perceptual tasks (e.g. contrast detection, Vernier, motion direction, and auditory frequency discrimination tasks) while high-level tasks were more immune
to training sequence (e.g. shape search task). Experiment 5 investigated the role of task difficulty, internal and external feedback in perceptual learning in a 4-Alternative Forced-Choice (4-AFC) Gabor orientation discrimination task. We found that difficult tasks mainly showed between-session forgetting while easy tasks were more subject to within-session adaptation. Adding internal feedback induced by easy task (i.e., 70% correct) or complete external feedback improved initial performance while providing external feedback speeded up general learning rate, introduced relearning process and modulated between-session effects. In Experiment 6, we co-trained two different tasks, e.g. face view discrimination and Gabor orientation discrimination, with two different performance levels (e.g. 35% and 70%) in four groups. Compared with mixed training of two tasks with same accuracies (e.g., 35% face + 35% orientation), mixed training with different accuracies (35% face + 70% orientation) led to worse initial performance or introduced within-session adaptation, but spared learning rate.
The current study proposed a comprehensive framework for the multi-component mechanism of perceptual learning. Applying the model to analyze a series of behavioral data demonstrated the rationality of our three hypotheses and identified learning mechanisms underlying different experimental settings. Our model provides a new perspective to understand the fundamental question of how behavioral benefits accumulated as training proceeds and an analytical tool to accommodate possible memory effects involved in learning across a variety of tasks and experimental conditions. In addition, the current research provides strong implications for a flexible and hybrid learning mechanisms and shows the great potential of utilizing the multi-component framework to optimize learning outcomes in real-world applications. |
修改评论