Disentangling locus of perceptual learning in the visual hierarchy of motion processing

Ru‐Yuan Zhang, Duje Tadin

TL;DR

This study investigates where in the visual system perceptual learning for motion processing occurs. By training participants on either component or pattern motion stimuli and testing for transfer, they found that learning primarily affects intermediate visual areas responsible for integrating component motion into a coherent pattern, rather than early-stage processing. This learning is also specific to trained stimulus features, ruling out broad, non-sensory decision-making improvements.

What Problem Does This Paper Address?

Visual perceptual learning (VPL) is the phenomenon where repeated exposure to a visual task leads to long-lasting improvements in performance. A key question in VPL research is where in the brain's visual processing hierarchy these improvements take place. The visual system is organized hierarchically, with early areas processing basic features (like edges and simple motion directions) and later areas integrating these into more complex representations (like object shapes and perceived motion of complex patterns). Understanding the locus of VPL is crucial for developing theories of learning and plasticity. In motion processing, there's a known transition from early visual areas (like V1) that respond to component motion (e.g., the directions of individual lines in a plaid) to middle-level areas (like MT/V3A) that respond to the overall perceived pattern motion. This paper aims to pinpoint whether VPL in motion perception occurs at these early component-processing stages or at the later pattern-processing stages.

Methods

The researchers employed a clever experimental design to disentangle learning at different visual processing levels. They trained two groups of participants on a motion direction identification task: one group trained on simple moving gratings (component motion) and the other on plaid stimuli (pattern motion). Crucially, they used plaid stimuli composed of two component gratings moving in different directions, which are perceived as a single, coherent moving object.

After training, participants were tested on various stimulus conditions, including the trained stimulus, stimuli with different component/pattern motion directions, and stimuli with variations in speed, size, and contrast. By observing how performance improved (or didn't improve) across these different conditions, they could infer the locus of learning.

  • Training: Two-alternative forced-choice motion direction identification task.
  • Stimuli: Moving gratings (component motion) and plaid stimuli (pattern motion).
  • Testing: Pre- and post-training assessments using a range of stimuli to measure transfer of learning.
  • Analysis: Duration thresholds were measured using interleaved staircases. Learning was quantified as "percent of improvement" (PI). Statistical comparisons (paired t-tests, one-sample t-tests) were used to assess significance.

Datasets

Not applicable. This study uses human participants and controlled visual stimuli, not pre-existing datasets.

Key Insights and Ideas

  • Bidirectional Transfer Points to Middle-Level Processing: The most significant finding is the complete, bidirectional transfer of learning between component and pattern motion stimuli when they share the same perceived direction. This suggests that learning isn't confined to the early stages that process individual components, nor is it solely at a high-level decision stage. Instead, it implicates the intermediate stages where component motions are integrated into a coherent pattern perception.
  • Specificity Rules Out Non-Sensory Learning: The observed specificity of learning to the trained stimulus direction, speed, size, and contrast strongly argues against broad, non-sensory learning mechanisms (like changes in task rules or decision strategies). If learning were purely conceptual or decision-based, it would be expected to transfer more broadly across variations in low-level features.
  • Plaid Perception as a Window into Hierarchy: The use of plaid stimuli, which exploit the known dissociation between component and pattern motion processing in the visual hierarchy, was key to disentangling the locus of learning. This design allowed them to psychophysically infer neural changes.

Key Results

  • Significant Perceptual Learning: Both the component-training and plaid-training groups showed significant improvements in their trained task, with approximately a 20% increase in performance (PI).
  • Complete Transfer Between Component and Pattern Motion:
    • Training on component motion (gratings) significantly improved performance on plaid stimuli moving in the same direction. The degree of improvement (PI) was statistically equivalent to the improvement on the trained grating.
  • Conversely, training on plaid motion significantly improved performance on component gratings that moved in the same perceived direction. Again, the PI was not statistically different from the PI on the trained plaid.
  • Lack of Transfer to Different Directions: Learning did not transfer to plaid stimuli that included the trained component but moved in a different direction. Similarly, plaid training did not improve performance on its constituent components when they moved in different directions.
  • Specificity to Stimulus Features: Learning showed significant specificity to the trained stimulus direction, speed, size, and contrast. Transfer was minimal or non-existent when these features were altered.

Limitations and Future Work

  • Stimulus Ambiguity: The authors acknowledge that their stimuli did not allow them to fully disentangle pattern motion perception from other potential mechanisms, such as tracking moving intersections within the plaid.
  • "Readout" Mechanisms: While the results argue against broad non-sensory learning, they cannot entirely rule out stimulus-selective changes in "sensory readout" mechanisms, which could still be sensitive to basic stimulus features.
  • Future Directions: The paper suggests that future work could explore the neural underpinnings of these findings using neuroimaging techniques (like fMRI) to directly observe activity in areas like MT/V3A during these learning paradigms. Further research could also investigate how attention and task relevance interact with these learning mechanisms. The authors also note that while their findings argue against broad non-sensory factors, the precise interplay of sensory and cognitive factors in VPL remains an open question.

Summary updated

Abstract

Visual perceptual learning (VPL) can lead to long-lasting perceptual improvements. One of the central topics in VPL studies is the locus of plasticity in the visual processing hierarchy. Here, we tackled this question in the context of motion processing. We took advantage of an established transition from component-dependent representations at the earliest level to pattern-dependent representations at the middle-level of cortical motion processing. Two groups of participants were trained on the same motion direction identification task using either grating or plaid stimuli. A set of pre- and post-training tests was used to determine the degree of learning specificity and generalizability. This approach allowed us to disentangle contributions from different levels of processing stages to behavioral improvements. We observed a complete bi-directional transfer of learning between component and pattern stimuli that moved to the same directions, indicating learning-induced plasticity associated with intermediate levels of motion processing. Moreover, we found that motion VPL is specific to the trained stimulus direction, speed, size, and contrast, diminishing the possibility of non-sensory decision-level enhancements. Taken together, these results indicate that, at least for the type of stimuli and the task used here, motion VPL most likely alters visual computation associated with signals at the middle stage of motion processing.
BibTeX
@inproceedings{zhang2019disentangling,
  author = {Ru‐Yuan Zhang and Duje Tadin},
  title = {{Disentangling locus of perceptual learning in the visual hierarchy of motion processing}},
  booktitle = {Scientific},
  year = {2019},
  doi = {10.1038/s41598-018-37892-x},
}
Figure