Novel representations that support rule-based categorization are acquired on-the-fly during category learning

Abstract

Humans learn categorization rules that are aligned with separable dimensions through a rulebased learning system, which makes learning faster and easier to generalize than categorization rules that require integration of information from different dimensions. Recent research suggests that learning to categorize objects along a completely novel dimension changes its perceptual representation, making it more separable and discriminable. Here we asked whether such newly learned dimensions could support rule-based category learning. One group received extensive categorization training and a second group did not receive such training. Later, both groups were trained in a task that made use of the category-relevant dimension, and then tested in an analogical transfer task (Experiment 1) and a button-switch interference task (Experiment 2). We expected that only the group with extensive pre-training (with well-learned dimensional representations) would show evidence of rule-based behavior in these tasks. Surprisingly, both groups performed as expected from rule-based learning. A third experiment tested whether a single session (less than one hour) of training in a categorization task would facilitate learning in a task requiring executive function. There was a substantial learning advantage for a group with brief pre-training with the relevant dimension. We hypothesize that extensive experience with separable dimensions is not required for rule-based category learning; rather, the rulebased system may learn representations "on the fly" that allow rule application. We discuss what kind of neurocomputational model might explain these data best.

Absence of cross-modality analogical transfer in perceptual categorization

“…In pilot work3, we found that the strategies participants reported using differed from what we predicted. In previous studies, the majority of participants used the optimum diagonal strategy in the II task throughout the experiment (Casale et al, 2012;Soto & Ashby, 2019;Zakrzewski et al, 2018). In contrast, we found that the majority of participants reported using twodimensional rules at training and one-dimensional rules at test.…”

“…However, those who learned an II structure could not. In a similar study that used rectangles of visual static (random green dots) that varied in dot density and rectangle size, Zakrzewski, Church, and Smith (2018) found the same pattern (see also Soto & Ashby, 2019). In summary, previous studies indicate that analogical transfer is possible for UD tasks but not for II tasks.…”

“…Soto & Ashby, 2019;Zakrzewski et al, 2018). They found that successful analogical transfer was contingent on the category structure: transfer was easier following UD training than II training.…”

Changes within neural population codes can be inferred from psychophysical threshold studies

“…For example, face features are thought to be encoded through monotonic tuning functions (e.g., sigmoidal; see [5,45,43]). Using computational modeling and visual adaptation, it has been found that the effects of categorization on perception of face identities along the category-relevant dimension [59,60,61] can be best explained using a specific gain mechanism [45]. It is currently unknown exactly how the complex shape and object stimuli used in some studies are encoded, but encoding that is different from that of orientation might be at the heart of the results obtained with such dimensions.…”

Adaptation aftereffects reveal how categorization training changes the encoding of face identity

“…Category learning is usually faster and more generalizable when the category bound aligns with a pre-existing dimensional structure in the stimuli. Morphed face dimensions lack such dimensional structure, but they appear to acquire it on-the-fly during training in a categorization task (Soto & Ashby, 2019). The dimension enhancement hypothesis would explain why such learning of representations that support categorization is so fast, as it would require only the modification of already-existing representations rather than the creation of new representations.…”