People can become more sensitive to small changes in what they are seeing – such as detecting a slight change in the angle of a particular line – with practice. This process is called perceptual learning, but the improvement is often specific such that it is typically lost if the line moves to a new place, or a different line angle is used. Previous work does show that it is possible to transfer the learning to a new location or angle if the individual also practices another, seemingly irrelevant, task at the same or a later time – such as judging how bright the line is.

To understand what might be happening to produce these seemingly conflicting results, Xiong et al. used a technique called “continuous flash suppression” with human volunteers. This approach meant that the volunteers were shown an object (such as an angled line) in one eye, while their other eye viewed white noise similar to the “snowflakes” seen on an old-fashioned un-tuned television screen. The flashing snowflakes in one eye meant that the volunteers were not consciously aware of the presence of the angled line in the other eye.

The experiments revealed that perceptual learning at the new location or line angle happened when a subconsciously-observed object was shown in the new location or angle, or when the volunteers were asked to pay attention to the “subconscious object” when no object was actually there. This suggests that perceptual learning can happen in new conditions both through ‘bottom-up’ processes, which rely entirely on information coming in from the senses, and ‘top-down’ processes, which are influenced by what a person is aware of and paying attetion to. What is more, the results suggest that the classical observations of specificity in perceptual learning are likely to be a result of the lack of bottom-up and top-down influences in the untrained condition, when the volunteers work hard to improve their performance with the trained condition.

Future studies could directly look at what is going on in the brain when perceptual learning becomes less specific, for example by using a technique like functional magnetic resonance imaging to measure brain activity.

Visual perceptual learning is the process in which the observers improve their discrimination of fine differences of basic visual features, such as contrast, orientation, motion direction, etc., through practice. For several decades visual perceptual learning has been regarded as a distinct format of learning because it is specific to the orientation and retinal location of the trained stimulus (Schoups et al., 1995; Ahissar and Hochstein, 1997; Shiu and Pashler, 1992; Dosher and Lu, 1998; Poggio et al., 1992; Fiorentini and Berardi, 1980; Yu et al., 2004). Such learning specificity has inspired theories that interpret visual perceptual learning as a result of training induced neural plasticity in the early visual areas (Schoups et al., 1995; Karni and Sagi, 1991; Teich and Qian, 2003; Bejjanki et al., 2011). For example, it has been proposed that training could sharpen neuronal orientation tuning in the primary visual cortex (V1), so that neurons become more sensitive to the fine changes of orientation differences (Teich and Qian, 2003; Schoups et al., 2001). The learning specificity has also constrained alternative reweighting theories of visual perceptual learning (Dosher and Lu, 1998; Poggio et al., 1992; Yu et al., 2004; Mollon and Danilova, 1996; Petrov et al., 2005; Law and Gold, 2008; 2009). These theories propose that training may not change the tuning properties of sensory neurons. Rather, the inputs from activated neurons are reweighted through training to improve readout at a later decision stage. Reweighting theories are supported by neurophysiological evidence. For example, motion direction learning in monkeys is correlated with changes in motion-driven responses of neurons in the lateral intraparietal area (LIP) that is related to the transformation of motion information into decisions (saccadic choices), but not with changes of neurons in the medial temporal area (MT) that represents motion direction signals (Law and Gold, 2008).

However, in a series of 'double training' studies we demonstrated that visual perceptual learning of various tasks can transfer significantly, and often completely, to new orientations or locations (Xiao et al., 2008; Wang et al., 2012, 2014; Zhang et al., 2010, 2014). In a double training experimental design the observers are additionally exposed to the new orientation or location via practicing an irrelevant task besides the primary learning task. For example, perceptual learning of foveal orientation discrimination (e.g., which of two consecutively presented gratings is more clockwise-tilted?) initially shows little transfer to an orthogonal orientation. However, if the observers are exposed to an orthogonal orientation via an irrelevant contrast discrimination task (e.g., which of two gratings has higher contrast?), learning transfers completely to the orthogonal orientation (Zhang et al., 2010). Similarly, perceptual learning of Vernier discrimination (e.g., whether a lower grating is placed to the left or right of an upper grating) can also transfer significantly and often completely to a new retinal location when the observers are additionally exposed to the transfer location via an irrelevant contrast or orientation discrimination task (Wang et al., 2012, 2014). These results suggest two important insights regarding visual perceptual learning. First, visual perceptual learning is mainly a high-level rule-based learning process that occurs beyond the retinotopic and orientation selective visual areas, so that learning is in principle transferrable to untrained conditions (Zhang et al., 2010). Second, learning specificity may be related to the untrained conditions, rather than the trained conditions as the field has been assuming (i.e., plasticity with the trained early visual cortical neurons or reweighting of the inputs from these neurons). The second insight, which stands completely different from the interpretations of specificity by the field, forms the basis of the current study.

During training at a specific orientation or location in a typical visual perceptual learning experiment, most mental resources are devoted to the trained stimuli. For example, an observer has to focus the attention on the near-threshold difference between the reference orientation and the target orientation at a specific location. As a result the untrained orientations and locations, and thus the visual neurons representing these orientations and locations, are neither bottom-up stimulated nor top-down attended during training. We thus suspect that this insufficient bottom-up stimulation of, and/or top-down attention to, the untrained conditions are responsible for the orientation and location specificity. In other words, because visual neurons representing untrained conditions are not properly activated as a result, high-level perceptual learning may not be able to functionally connect to these neurons for learning transfer.

Our previous double training experiments are unable to separately identify the potential bottom-up and/or top-down contributions because suprathreshold stimuli are used in the secondary exposure task. In the current study we applied a continuous flash suppression (CFS) technique (Tsuchiya and Koch, 2005) to suppress the exposure stimulus into sub-consciousness (see Materials and methods). We further manipulated the subconscious stimulus conditions to make the exposure task to be bottom-up only or top-down only. The results show that either bottom-up stimulation of the untrained condition, or top-down attention to it, is sufficient to enable substantial and often complete transfer of learning. These results provide a solution to the mystery of learning specificity that has dominated the history of perceptual learning research. With learning specificity considered as a by-product of training, the field should move on to study the brain mechanisms of perceptual learning without much of specificity-related constraints. Moreover, more efficient training paradigms can be designed to generate perceptual learning without the unwanted specificity in practical settings.

A major focus of perceptual learning research has been on the links between learning specificity and training-induced neural changes in the brain. The common explanations include learning specificity as a product of neural plasticity in early visual areas that are most feature selective and retinotopic (Karni and Sagi, 1991; Teich and Qian, 2003; Schoups et al., 2001), or of improved readout of inputs from early visual areas (Dosher and Lu, 1998; Poggio et al., 1992; Mollon and Danilova, 1996; Law and Gold, 2009). Here our results paint a completely opposite picture: It is the actions (or the absence of actions) with the untrained conditions that decide the learning specificity and transfer. We show that for orientation and location specificity, the absence of bottom-up stimulation of neurons representing the untrained conditions, as well as top-down attention to these neurons, prevent perceptual learning from transferring to untrained conditions.

In one ERP study (Zhang et al., 2013), we discovered that Vernier learning and its transfer to an untrained hemisphere accompanies significant occipital P1-N1 changes when the Vernier task is performed at either the trained or the untrained location after training. However, if learning does not transfer, as shown in about half the observers, the P1-N1 changes are limited to the trained location. We interpret P1-N1 changes as possible indications of top-down connections between high-level Vernier learning and visual neurons at the trained location, as well as at untrained locations when learning transfers. We also interpret learning specificity as a result of absent functional connections. Our current data suggest that both bottom-up stimulation of, and top-down attention to, untrained conditions may activate visual neurons representing the untrained conditions, so as to foster the connections to enable learning transfer.

In another ERP study (Zhang et al., 2015), we also found that orientation discrimination learning and its transfer to an untrained hemisphere accompanies significant occipital C1 changes when the orientation discrimination task is performed at the trained and untrained locations, respectively, after training (Zhang et al., 2015). Such C1 changes are not evident with an untrained shape-discrimination task (Zhang et al., 2015) or with passive viewing (Bao et al., 2010). These results further indicate that the functional connections are task specific, consistent with the observations that perceptual learning is task specific (Shiu and Pashler, 1992; Ahissar and Hochstein, 1993; Cong et al., 2016). The task-specific functional connections associated with learning and its transfer are also supported by fMRI evidence, in that the generalized orientation discrimination learning is accompanied with task-specific enhancement of orientation-selective responses in the early visual areas including V1, V2 and V3 (Byers and Serences, 2014). These results together may also explain fMRI results that orientation specificity is accompanied with refined representation of the trained orientation in early visual areas (Jehee et al., 2012). This is because top-down modulation by high-level orientation learning may not reach the early cortical representations of untrained orientations as a result of absent functional connections.

As we pointed out earlier, when a near-threshold task is practiced, most brain resources are devoted to the training orientation and retinal location. The untrained orientations and retinal locations, and thus the relevant visual neurons, are neither bottom-up stimulated nor top-down attended. At least in the case of location specificity, there is evidence that spatial attention could suppress other unattended locations even with no presence of competing stimuli (Smith et al., 2000; Shmuel et al., 2006), as is typical in a perceptual learning study. We suspect that some or all of these factors could contribute to under-activations of neurons at untrained conditions, which in turn could lead to missing functional connections of these neurons to high-level learning to prevent learning transfer.

Our previous double training studies have revealed significant and often complete learning transfer to untrained conditions (Xiao et al., 2008; Zhang et al., 2010). These transfer results prompted us to propose a rule-based perceptual learning theory. Specifically, we suggest that visual perceptual learning is a high-level process. The brain learns the rules of reweighting visual inputs to achieve better visual performance, and these rules are potentially applicable to new orientations and retinal locations. More recent evidence also suggests that these rules are conceptual, in that learning can transfer between physically distinct stimuli that are initially encoded by different neural mechanisms (e.g., between local and global orientations defined by gratings and symmetric dot patterns, or between first- and second-order motion directions) (Wang et al., 2016). The current findings add to this theory by elucidating why perceptual learning, if high-level and rule-based, can be specific in the first place. These results also suggest that double training schemes function by providing bottom-up and top-down forces to activate untrained neurons, which in turn initiate functional connections for learning transfer. A recent computational model explains how such functional connections can be built with double training (Solgi et al., 2013). In the model, the secondary training task activates the untrained neurons, which could be recalled when the brain is off-task, so that high-level 'concept neurons' that have learned the task can connect to these untrained neurons in an off-task self-organization manner for learning transfer. Apparently the activations of untrained neurons due to unconsciousness stimulation, or top-down attention without actual stimulation, can also be recalled off-task in the context of this computational account.

We once had observers practice a peripheral Vernier task identical to the current one, while flashing a Gabor simultaneously at a diagonal location (without noise suppression) for the purpose of stimulating neurons at this latter location (Wang et al., 2012). However, we found no evidence for learning transfer. A more recent psychophysical study also replicated the null-transfer results (Mastropasqua et al., 2015). Similarly, an earlier monkey-recording study (Schoups et al., 2001) reported that orientation discrimination learning does not transfer to a different location where a same grating stimulus is simultaneously presented. These null-transfer effects may be caused by attentional competition, in that concentrated attention to the training location would suppress the high-contrast stimuli at a different location (Watanabe et al., 2001). The attentional suppression effect is avoided in double training when the primary training and the secondary training that stimulates the untrained location are performed in alternating blocks of trials (Xiao et al., 2008), or when the stimulation of the untrained locations is below awareness as in our current study. In the latter case the bottom-up stimulation of transfer location through one eye may create an input contrast in the eye-of-origin feature, and this contrast, although invisible perceptually, has been shown to be very salient because it more strongly activates V1 neurons than the surrounding stimuli (Zhaoping, 2008).

Perceptual learning can also be task irrelevant. Watanabe and colleagues (Watanabe et al., 2001; Seitz and Watanabe, 2003) reported that training improves discrimination of a nearby feature that is task irrelevant and sub-threshold (to avoid attentional suppression), and that the learning occurs only when the irrelevant feature is temporally paired with rewards with the trained task. They explained this task irrelevant perceptual learning (TIPL) as a result of interactions between spatially diffusive rewards and bottom-up exposure of the task irrelevant feature (Seitz and Watanabe, 2005). Note that our double-training enabled learning transfer are distinct from TIPL in two important ways. First, unlike TIPL, the training and exposure do not require temporal pairing and spatial proximity of two stimuli. Indeed the training phase can precede the exposure phase with a time gap of up to 8 weeks (Zhang et al., 2010), and the training and exposure locations can be at diagonal quadrants of the visual field separated by 10 degrees (Wang et al., 2012), while double training is still effective. In the current study, the training and sub-threshold exposure are performed in separate blocks of trials, rather than paired, in Figure 1, and are spatially separated by 10 degrees in Figure 4. Second, our recent evidence suggests that double-training enabled learning transfer is still task specific (Cong et al., 2016). We found that orientation discrimination learning cannot transfer to a contrast discrimination task using the same Gabor stimulus, even after the observers receive additional exposure of the transfer task through easy trials in separate blocks of trials. The same is true at a reverse direction when the observers learn contrast discrimination and receive exposure of orientation discrimination, the transfer task, through easy trials.

1. 1. Ahissar M
   2. Hochstein S

   (1993) Attentional control of early perceptual learning

   PNAS 90:5718–5722.

   https://doi.org/10.1073/pnas.90.12.5718

   • Google Scholar
2. 1. Ahissar M
   2. Hochstein S

   (1997) Task difficulty and the specificity of perceptual learning

   Nature 387:401–406.

   https://doi.org/10.1038/387401a0

   • Google Scholar
3. 1. Bao M
   2. Yang L
   3. Rios C
   4. He B
   5. Engel SA

   (2010) Perceptual learning increases the strength of the earliest signals in visual cortex

   Journal of Neuroscience 30:15080–15084.

   https://doi.org/10.1523/JNEUROSCI.5703-09.2010

   • Google Scholar
4. 1. Bejjanki VR
   2. Beck JM
   3. Lu ZL
   4. Pouget A

   (2011) Perceptual learning as improved probabilistic inference in early sensory areas

   Nature Neuroscience 14:642–648.

   https://doi.org/10.1038/nn.2796

   • Google Scholar
5. 1. Byers A
   2. Serences JT

   (2014) Enhanced attentional gain as a mechanism for generalized perceptual learning in human visual cortex

   Journal of Neurophysiology 112:1217–1227.

   https://doi.org/10.1152/jn.00353.2014

   • Google Scholar
6. 1. Cong LJ
   2. Wang RJ
   3. Yu C
   4. Zhang JY

   (2016) Perceptual learning of basic visual features remains task specific with Training-Plus-Exposure (TPE) training

   Journal of Vision 16:13.

   https://doi.org/10.1167/16.3.13

   • Google Scholar
7. 1. Dosher BA
   2. Lu Z-L

   (1998) Perceptual learning reflects external noise filtering and internal noise reduction through channel reweighting

   PNAS 95:13988–13993.

   https://doi.org/10.1073/pnas.95.23.13988

   • Google Scholar
8. 1. Fiorentini A
   2. Berardi N

   (1980) Perceptual learning specific for orientation and spatial frequency

   Nature 287:43–44.

   https://doi.org/10.1038/287043a0

   • Google Scholar
9. 1. Jehee JF
   2. Ling S
   3. Swisher JD
   4. van Bergen RS
   5. Tong F

   (2012) Perceptual learning selectively refines orientation representations in early visual cortex

   Journal of Neuroscience 32:16747–16753.

   https://doi.org/10.1523/JNEUROSCI.6112-11.2012

   • Google Scholar
10. 1. Karni A
    2. Sagi D

    (1991) Where practice makes perfect in texture discrimination: evidence for primary visual cortex plasticity

    PNAS 88:4966–4970.

    https://doi.org/10.1073/pnas.88.11.4966

    • Google Scholar
11. 1. Law CT
    2. Gold JI

    (2008) Neural correlates of perceptual learning in a sensory-motor, but not a sensory, cortical area

    Nature Neuroscience 11:505–513.

    https://doi.org/10.1038/nn2070

    • Google Scholar
12. 1. Law CT
    2. Gold JI

    (2009) Reinforcement learning can account for associative and perceptual learning on a visual-decision task

    Nature Neuroscience 12:655–663.

    https://doi.org/10.1038/nn.2304

    • Google Scholar
13. 1. Mastropasqua T
    2. Galliussi J
    3. Pascucci D
    4. Turatto M

    (2015) Location transfer of perceptual learning: passive stimulation and double training

    Vision Research 108:93–102.

    https://doi.org/10.1016/j.visres.2015.01.024

    • Google Scholar
14. 1. Mollon JD
    2. Danilova MV

    (1996) Three remarks on perceptual learning

    Spatial Vision 10:51–58.

    https://doi.org/10.1163/156856896X00051

    • Google Scholar
15. 1. Petrov AA
    2. Dosher BA
    3. Lu ZL

    (2005) The dynamics of perceptual learning: an incremental reweighting model

    Psychological Review 112:715–743.

    https://doi.org/10.1037/0033-295X.112.4.715

    • Google Scholar
16. 1. Poggio T
    2. Fahle M
    3. Edelman S

    (1992) Fast perceptual learning in visual hyperacuity

    Science 256:1018–1021.

    https://doi.org/10.1126/science.1589770

    • Google Scholar
17. 1. Schoups A
    2. Vogels R
    3. Qian N
    4. Orban G

    (2001) Practising orientation identification improves orientation coding in V1 neurons

    Nature 412:549–553.

    https://doi.org/10.1038/35087601

    • Google Scholar
18. 1. Schoups AA
    2. Vogels R
    3. Orban GA

    (1995) Human perceptual learning in identifying the oblique orientation: retinotopy, orientation specificity and monocularity

    Journal of Physiology 483:797–810.

    https://doi.org/10.1113/jphysiol.1995.sp020623

    • Google Scholar
19. 1. Seitz A
    2. Watanabe T

    (2005) A unified model for perceptual learning

    Trends in Cognitive Sciences 9:329–334.

    https://doi.org/10.1016/j.tics.2005.05.010

    • Google Scholar
20. 1. Seitz AR
    2. Watanabe T

    (2003) Psychophysics: Is subliminal learning really passive?

    Nature 422:36.

    https://doi.org/10.1038/422036a

    • Google Scholar
21. 1. Shiu LP
    2. Pashler H

    (1992) Improvement in line orientation discrimination is retinally local but dependent on cognitive set

    Perception & Psychophysics 52:582–588.

    https://doi.org/10.3758/BF03206720

    • Google Scholar
22. 1. Shmuel A
    2. Augath M
    3. Oeltermann A
    4. Logothetis NK

    (2006) Negative functional MRI response correlates with decreases in neuronal activity in monkey visual area V1

    Nature Neuroscience 9:569–577.

    https://doi.org/10.1038/nn1675

    • Google Scholar
23. 1. Smith AT
    2. Singh KD
    3. Greenlee MW

    (2000) Attentional suppression of activity in the human visual cortex

    NeuroReport 11:271–277.

    https://doi.org/10.1097/00001756-200002070-00010

    • Google Scholar
24. 1. Solgi M
    2. Liu T
    3. Weng J

    (2013) A computational developmental model for specificity and transfer in perceptual learning

    Journal of Vision 13:7.

    https://doi.org/10.1167/13.1.7

    • Google Scholar
25. 1. Teich AF
    2. Qian N

    (2003) Learning and adaptation in a recurrent model of V1 orientation selectivity

    Journal of Neurophysiology 89:2086–2100.

    https://doi.org/10.1152/jn.00970.2002

    • Google Scholar
26. 1. Tsuchiya N
    2. Koch C

    (2005) Continuous flash suppression reduces negative afterimages

    Nature Neuroscience 8:1096–1101.

    https://doi.org/10.1038/nn1500

    • Google Scholar
27. 1. Wang R
    2. Wang J
    3. Zhang JY
    4. Xie XY
    5. Yang YX
    6. Luo SH
    7. Yu C
    8. Li W

    (2016) Perceptual Learning at a Conceptual Level

    Journal of Neuroscience 36:2238–2246.

    https://doi.org/10.1523/JNEUROSCI.2732-15.2016

    • Google Scholar
28. 1. Wang R
    2. Zhang JY
    3. Klein SA
    4. Levi DM
    5. Yu C

    (2012) Task relevancy and demand modulate double-training enabled transfer of perceptual learning

    Vision Research 61:33–38.

    https://doi.org/10.1016/j.visres.2011.07.019

    • Google Scholar
29. 1. Wang R
    2. Zhang JY
    3. Klein SA
    4. Levi DM
    5. Yu C

    (2014) Vernier perceptual learning transfers to completely untrained retinal locations after double training: a "piggybacking" effect

    Journal of Vision 14:12.

    https://doi.org/10.1167/14.13.12

    • Google Scholar
30. 1. Watanabe T
    2. Náñez JE
    3. Sasaki Y

    (2001) Perceptual learning without perception

    Nature 413:844–848.

    https://doi.org/10.1038/35101601

    • Google Scholar
31. 1. Xiao LQ
    2. Zhang JY
    3. Wang R
    4. Klein SA
    5. Levi DM
    6. Yu C

    (2008) Complete transfer of perceptual learning across retinal locations enabled by double training

    Current Biology 18:1922–1926.

    https://doi.org/10.1016/j.cub.2008.10.030

    • Google Scholar
32. 1. Yu C
    2. Klein SA
    3. Levi DM

    (2004) Perceptual learning in contrast discrimination and the (minimal) role of context

    Journal of Vision 4:169–182.

    https://doi.org/10.1167/4.3.4

    • Google Scholar
33. 1. Zhang GL
    2. Cong LJ
    3. Song Y
    4. Yu C

    (2013) ERP P1-N1 changes associated with Vernier perceptual learning and its location specificity and transfer

    Journal of Vision 13:19.

    https://doi.org/10.1167/13.4.19

    • Google Scholar
34. 1. Zhang GL
    2. Li H
    3. Song Y
    4. Yu C

    (2015) ERP C1 is top-down modulated by orientation perceptual learning

    Journal of Vision 15:8.

    https://doi.org/10.1167/15.10.8

    • Google Scholar
35. 1. Zhang J-Y
    2. Cong L-J
    3. Klein SA
    4. Levi DM
    5. Yu C

    (2014) Perceptual learning improves adult amblyopic vision through rule-based cognitive compensation

    Investigative Opthalmology & Visual Science 55:2020–2030.

    https://doi.org/10.1167/iovs.13-13739

    • Google Scholar
36. 1. Zhang JY
    2. Zhang GL
    3. Xiao LQ
    4. Klein SA
    5. Levi DM
    6. Yu C

    (2010) Rule-based learning explains visual perceptual learning and its specificity and transfer

    Journal of Neuroscience 30:12323–12328.

    https://doi.org/10.1523/JNEUROSCI.0704-10.2010

    • Google Scholar
37. 1. Zhang T
    2. Xiao LQ
    3. Klein SA
    4. Levi DM
    5. Yu C

    (2010) Decoupling location specificity from perceptual learning of orientation discrimination

    Vision Research 50:368–374.

    https://doi.org/10.1016/j.visres.2009.08.024

    • Google Scholar
38. 1. Zhaoping L

    (2008) Attention capture by eye of origin singletons even without awareness--a hallmark of a bottom-up saliency map in the primary visual cortex

    Journal of Vision 8:1.

    https://doi.org/10.1167/8.5.1

    • Google Scholar

Author details

1. Ying-Zi Xiong

   School of Psychological and Cognitive Sciences, Peking University, Beijing, China

   Contribution

   Y-ZX, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

2. Jun-Yun Zhang

   School of Psychological and Cognitive Sciences, Peking University, Beijing, China

   Contribution

   J-YZ, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

3. Cong Yu

   1. School of Psychological and Cognitive Sciences, Peking University, Beijing, China
   2. IDG/McGovern Institute for Brain Research, Peking University, Beijing, China
   3. Peking-Tsinghua Center for Life Sciences, Peking University, Beijing, China

   Contribution

   CY, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   yucong@pku.edu.cn

   Competing interests

   The authors declare that no competing interests exist.

   "This ORCID iD identifies the author of this article:" 0000-0002-8453-6974

• 2,048

  views

• 353

  downloads

• 26

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.