We are constantly exposed to an overwhelming amount of sensory signals, most of which are not noteworthy. To function normally in the world, we must react quickly and precisely to the important signals, while ignoring or discounting the less important, just as organisms must do in the natural environment to survive. By directing attention only to important signals or being repeatedly exposed to signals in an important context, our sensory systems learn to process important signals more efficiently than the less-important signals. Reflecting this fundamental role of perceptual learning, studies have been conducted to examine mechanisms of perceptual learning and neural plasticity with various kinds of tasks and stimuli by using behavioral measurements [2,3] and neurophysiological [7,8] and brain imaging techniques [9,10]. Perceptual learning and neural plasticity have also been studied in all the sensory modalities including vision, hearing [11], and touch perception [12].

(A) Early stage, local network model: Adini et al. [13] and Tsodyks, et al. [14] have proposed the model based on perceptual learning of contrast discrimination (indicating whether two contrast values are the same or different). In their model, repeated presentation of a stimulus (a line presented in the center) together with its surrounding stimulus (lines presented around the central line) leads to a change in the visual cortex as a result of interactions between signals from the central and surrounding stimuli. That is, for an observer to learn a stimulus in one location (the central stimulus), the context in which the stimulus is presented (surrounding stimuli) plays an important role. The interactions can occur within the same cortex and therefore we do not have to assume interactions between cortical areas at different levels in the visual processing hierarchy. In this model, the neural reorganization due to perceptual learning can occur in a low-level cortex, including the primary visual cortex (V1), which is the first visual cortex onto which visual signals are projected. This model indicates the mechanism of a type of perceptual learning that can involve only one level of visual processing and suggests that perceptual learning does not necessarily require lower-to-higher or higher-to-lower connections between different cortical areas at different stages of visual processing.

(B) Mid-level stage, reweighting: In the model by Dosher and Lu [15–17], learning occurs by changing the strength (reweighting) of neural connections between the early visual stage, such as V1, in which highly local processing occurs, and a decision unit. The changes occur in the neural connections specifically for a given task. In this sense, it is possible that different stages between the earliest visual stage and the decision unit are involved.

(C) Higher-to-lower stages: Reverse hierarchy theory (RHT) indicates that learning is an attention guided process [18–20] (Figure 2). According to this model, visual learning begins at high-level visual areas that may be able to deal with a task requiring discriminating signals with large differences (Figure 1). When a task requires discrimination of signals with smaller differences, the site of learning proceeds to lower visual areas where signals with smaller differences can be discriminated. RHT indicates that learning is driven by attention that selects neuronal population suitable for the levels of the signal.

Based on these findings, Zhang et al. have proposed the “stimulus tagging model” for perceptual learning. This model indicates that to learn multiple stimuli, the brain needs to conceptually “tag” each stimulus that differs in the feature value. A presentation of feature values in a structured way over time or a presentation of the same label paired with a specific value of a stimulus feature is regarded as tagging. For example, the letter “A” is paired with one feature value and the letter “B” with another feature value. Thus, tagging can be conceptual. Such tagging guides attention directly to the appropriate perceptual template.

How is the stimulus tagging model applied to real life situations? Imagine that a young animal is learning to visually discriminate multiple kinds of fruits. While all the fruits look alike, only one of them is edible. To learn to visually discriminate them, the animal could “tag” based on the different tastes of each fruit.

In the study by Zhang et al. [21], successful perceptual learning of multiple feature values occurs when the “tagging” successfully differentiated the feature values. For example, both presenting a letter “A” while discriminating contrasts around 0.2, and presenting a letter “B” while discriminating contrasts around 0.3 allowed the participants to learn to discriminate both contrasts around 0.2 and contrasts around 0.3. Since tagging was conceptual or semantic without needing to involve spatial information or conflicting signals, the type of attention involved in tagging is not likely to be either orienting that works for spatial information or executive control that deals with conflicting signals. The remaining type of attention is alertness. How does alertness play a role in perceptual learning with tagging?

Seitz and Watanabe have suggested that when a stimulus is presented concurrently with internal alertness elevation, perceptual learning of the stimulus occurs [26]. Perceptual learning with stimulus tagging might occur within this framework. When a previously presented temporal pattern or label is presented and successfully distinguished from other patterns or labels, the participant's alertness level might be elevated, and features presented concurrently with the pattern or label may be learned.

Clarifying the role of attention in perceptual learning provides clues to elucidate the underlying neural mechanisms of perceptual learning. Seitz and Watanabe [26] pointed out a potential link between perceptual learning and the alerting system by noting that the alerting system is associated with the right frontal regions and the right parietal regions [27-32]. For instance, lesions in the right frontal regions or the right parietal regions cause deficits in attention most related to alertness. In these areas, the norepinephrine system is thought to be involved in the activations [33,34]. On the other hand, neural activity in the locus coeruleus (LC), from where norepinephrine arises, is correlated with behavioral improvements on visual discrimination tasks [35]. A study has indicated that neural activities in the LC closely correlated with fluctuations in behavioral performance of visual discrimination task [36]. How perceptual learning and alerting attention are related, particularly in terms of the norepinephrine system, would be a highly interesting subject of future investigations.

Another potential future extension of the stimulus tagging model would be whether it can be generalized to other modalities. Although the tagging model is based on results of visual perceptual learning, it would be highly interesting to examine how well the tagging model fits with learning models associated with other sensory modalities, including audition and tactics, that have a hierarchical signal processing structure and top-down modulation just like the visual system.