Across two training phases participants learned about categories A, B, C, and D in Table 1 via inference or classification. Eye tracking was used throughout to monitor participants’ attention to the three feature dimensions and the category label. A test phase examined classification performance and attention profiles as people made novel category contrasts. From prior research we expected classification subjects to learn to ignore the irrelevant dimensions during training; this attention optimization should lead to a difficulty in making novel classifications. In contrast, prior research has demonstrated a tendency for inference learners to acquire within-category information, suggesting a general motivation to learn about all the dimensions in the inference task. Such motivation can potentially produce flexible category representations—that is—ones that support novel contrasts. Measuring eye movements during training will help explain differences in concept flexibility between groups. In contrast to previous studies comparing inference and classification, black plastic nursery pots a change was introduced to our inference training procedure: One of the dimensions, the contrast dimension 3, was never queried .
This change was made to better equate the two tasks; allowing inference participants to ignore task-irrelevant dimensions just like classification learners could. This allowed a test of whether inference learners are in fact generally motivated to learn about category features, or whether the demands of the task, i.e., querying the features, is what draws learners’ attention.Learning AB and CD training performance are shown in Figure 2. The figure shows average classification performance for the classification group and relevant cue inference performance. Both classification and inference groups improved over training blocks, but classification training was easier than inference training, with a higher proportion correct over blocks. The inference learners performed above chance levels in predicting the valid cue, t = 4.46, p < .01, but were marginally lower than the classification group, t = 1.81, p < .10 on the last AB training block. The CD training blocks were similar. Fixations A crucial question was whether inference learners fixated the non-queried dimension during learning. If inference is a more natural learning task than classification, it should motivate a general interest in learning about the category dimensions; fixations should be distributed to all dimensions, regardless of whether those dimensions are queried. However, if it is the attentional demands of the inference task that drive learning about dimensions , then fixations should shift away from the non-queried dimension, since it is no longer immediately relevant for the task.
The latter result would suggest that differences in what is learned via inference and classification are from different attentional requirements, and not motivational factors. Eye fixations will be used to distinguish between these two possibilities. Figure 3 shows proportion of fixations to category label and dimensions over AB and CD blocks, as a function of task. Replicating our earlier work, at the beginning of learning, the average classification learner fixated dimensions about equally. We also observed the expected shift in fixations from irrelevant to relevant dimensions, until irrelevant dimensions were fixated rarely or not at all. At the onset of CD training in block six, there is uneven attention distribution resulting from the learned fixation patterns from AB training, so that in the first trial of CD training, classification learners were not fixating the contrast dimension or the CD relevant dimension. A second attention optimization obtained for classification subjects. Recall that the contrast dimension was never queried. If inference motivates a general interest in the category features, we should observe continued fixations to the contrast dimension, in spite of it now being task-irrelevant. However, Figure 3 shows that throughout learning, inference learners largely ignored the contrast dimension. Although attention to dimensions 1 and 2 remained high throughout learning, even in the first learning block inference learners largely ignored the contrast dimension.
In fact, in the first block of learning, the amount of time fixating the contrast dimension was already significantly less than that of fixating the other two dimensions and the category label . Apparently, inference learners do in fact optimize their attention away from task-irrelevant cues. Attention optimization in the inference task contradicts the idea that inference motivates a general interest in the category features beyond what is strictly necessary. Rather, the results of Figure 3 support the idea that what distinguishes classification from inference is the attentional demand it places on the learner. Learners fixate dimensions because the task requires it and not because of motivational factors. Any motivation there may have been to learn about all of the category features extinguished quickly . Switch-trial performance Eye fixation data have ruled out that inference motivates general interest in category features. By not querying the contrast dimension in the inference condition, we allowed inference learners the opportunity to optimize their attention, just as the classification learners could. In fact, inference learners optimized their attention to just those queried dimensions, ignoring the never-queried contrast dimension. As a result of this manipulation, the inference learners may now struggle to include the contrast dimension, since they largely ignored it during training. On the other hand, although the inference learners never directed their attention to the contrast dimension, because it was not part of the task, they never had to learn to direct their attention away from that dimension either. Rather, the task focused their attention more on the two queried cues, and inference subject learned which dimensions were task-relevant. It is this fact that may still allow inference learning to nevertheless produce flexible attention allocation. By not learning to ignore the contrast dimension, inference learners may be free to use it during the switch trials. Blocks 10 and 11 of Figure 2 show proportion correct for switch-classification. In spite of not deploying significant fixations to the contrast dimension during training, the inference condition nevertheless showed an advantage during the switch trials. In the first block of switch trials, the inference group outperformed the classification group , t = 1.95, p = .064. Likewise, during the second block of switch trials, the inference group outperformed the classification group , t = 2.26, p < .05. Spending a large amount of time fixating a dimension during learning does not seem necessary for using that dimension later in a flexible way. Whatever inference subjects learned during training allowed them to perform well during switch trials. The eye movement results from training showed that classification and inference learners both largely ignored the contrast dimension. It makes sense then that classification learners should fail to use the contrast dimension during the switch trials, 30 plant pot but what allowed inference learners to have more flexible category representations than the classification group? Figure 4 shows learners’ attention allocation to the contrast dimension as a function of trial for the first block of AB and CD training . The figure shows that at the trial level, the largest attentional difference between the two conditions was that the classification learners allocated more attention to the contrast dimension early in learning. The different patterns of attention reflect different reasons the two groups probably ignored the contrast dimension. Inference learners ignored it because the task directed their attention to those dimensions being queried. Since the contrast dimension was never queried, their attention was never allocated to it. Classification learners were in a different position. From their perspective, any or all dimensions could have been important for getting the answer right, so they had to learn to ignore the contrast dimension, as they gradually discovered that the contrast dimension didn’t help them classify As from Bs or Cs from Ds.
We suspect that this is why there is an initial increase in fixations to the contrast dimension in the first CD block, because classification learners attended to it, and then learned that it was useless in classifying Cs and Ds. Classification learners’ fixation results reflected a learned inattention to the contrast dimension, which probably caused their difficulty in attending to the contrast dimension during the switch trials.We began with the observation that real-life categorizers can make novel category contrasts and that information learned about one set of categories transfers to another without difficulty. This observation seemed to be at odds with the robust finding that people in classification experiments tend to optimize their attention to the fewest necessary dimensions. Such optimization would necessarily force learners to reallocate attention when previously irrelevant dimensions at once become relevant. To resolve the contradiction that people can make novel category contrasts on one hand and but also tend to optimize attention on the other, we looked to other types of learning tasks they may produce classification performance that is less optimal but more flexible overall. Inference training seemed like the best candidate. There were two reasons for this. The first was based on evidence that inference yields a special type of processing in humans; although the exact source of this special processing was until now not entirely clear, classification learning has been found to cause humans to attend to diagnostic information and inference learning can cause learners to focus on within category correlations and prototypical features. We imagined that such differences may reflect that inference is a more typical learning task than classifying, and it isn’t hard to imagine how familiarity in the learning task can lead to greater ease and flexibility in using the acquired information. Our second hypothesis for how inference learning could yield flexible category representations was based on differences in attentional demands of inference and classification. Whereas most classification tasks allow learners to ignore some of the irrelevant dimensions, in the typical inference learning experiment, all of the dimensions are queried several times throughout training. Focusing people’s attention on all of the dimensions in this way may cause people to look at all dimensions on every trial, in order to prepare for future queries. In fact, the eye tracking results from this study show that never querying one of the dimensions allows the inference learner to optimize their attention to only those task-relevant dimensions, i.e., those dimensions that are sometimes queried. As it turned out, our initial hypotheses about inference learning were not exactly right. Our data showed that inference subjects very quickly ignored the never-queried dimension. Significant differences in fixations to the contrast dimension were found within the first learning block. Apparently, attending to the contrast dimension during training was not necessary for creating flexible category representations. Rather, what gave subjects the advantage in switch-classification trials is that they never had to learn to ignore the contrast dimension, as the classification subjects did, as evidence by the much larger drop in attention to the contrast dimension from the beginning to the end of training in the classification condition. In other words, classification subjects were harmed in their task by their learned attention profiles, but the inference subjects were not. Such a finding is in fact consistent with theories of attention and category learning. Several models, for example, RASHNL Kruschke and Johansen , and EXIT Kruschke , which are based on Macintosh’s theory of learned attention, propose that attention weights are learned for a given set of inputs. In these models, if feature inputs are irrelevant, or if for other reasons the features increase the number of classification errors committed, the attention system will direct attention away from those features in favor of others. These attention mechanisms help the models explain a large array of blocking and highlighting phenomena in addition to benchmark category learning data. They also explain why itis that our classification subjects failed to redirect attention to the contrast dimension during the switch trials. Beyond supporting certain theories of categorization and attention, our results underscore an important difference between the attention profiles acquired through trial and error learning and those that arise out of task goals. It seems that ignoring features as a result of discovering that they are statistically irrelevant over numerous trials is qualitatively different than cues that are never queried, and are thus irrelevant for the task. Thus, how the learner acquires an attention profile is as important as the attention profile itself.The 2020–2025 Dietary Guidelines for Americans encourages the intake of a variety of plant-based foods including nuts and berries. With the goal of increasing current knowledge on nuts and berries, as well as addressing research challenges and opportunities, the Nuts and Berries Conference: Pathways to Oxidant Defense, Vascular Function, and Gut Microbiome Changes was held on 5 to 6 May, 2022 at the University of California, Davis. Tree nuts and berries were selected as the focus of the conference for their unique composition, bioactivity, and multitude of associated health-promoting qualities.