Neural Mechanism Of Invariance Detection And Categorization Uncertainty In Category Induction

Category learning is usually defined by extracting common features among matters from discrete events, generating concepts and obtaining categorization for establishing response differently to objects or events. It is a vitally crucial skill for people in our daily life, and is a basic cognitive function in human brain.Previous studies used different category learning paradigms and technique methods to investigate the cognitive process and neural mechanism of category learning. Most of studies revealed that different category learning tasks had different cognitive processes and brain activities. For example, the requirement of explicit reasoning process was in rule-based category learning task, and it grasped one rule to learn certain categories. This category learning was related to the activations of prefrontal cortex and caudate. In information integration tasks, information from two or more stimuli needed to be integrated to learn features of correct categories. These features about categorization were difficult to be described verbally. Procedural memory was involved in the information integration task. This learning task activated base ganglia and lateral occipital region. In addition, prototype distortion task is to distort prototype stimuli. Subjects’ task is to extract the prototype category from these distorted stimuli. Prototype distortion learning primarily activated the prefrontal and parietal cortex, medial temporal lobe, and new striatum.Induction is the core ability in category learning. Category induction is to determine the relationship between two or more stimuli by comparing common features among stimuli. There are many studies about category induction on behavioral and EEG level. Chen et al. (2007,2008) divided category learning into three mental processes by using sophisticated segmentation design:category induction, categorization, cognitive control. They investigated these three processes by ERP technology. The results indicated that category induction effect is mainly reflected by the late positive component (LPC). The activation of the hippocampus and medial temporal lobe were related to category induction. The process of categorization triggered N2 and P3 components. The location of N2 component was in the anterior cingulate (ACC), which reflected the anticipation violation. The P3 was located in the medial temporal lobe, which is important for the long-time category process. The process of ignoring irrelevant information was closely related to the early N2 components in cognitive control. The N2 was located in ACC. In comparison with previous research about category learning, this study adopted ERP technology that is suitable for category learning paradigm, three processes of category induction, categorization and cognitive control were explored relatively independently. The study explored preliminarily neural mechanisms of artificial category learning.Previous studies investigated uncertainty of category induction on behavioral levels, and on issue such strategy adoption of uncertainty. It was less on investigating neural mechanism on aspects of uncertainty in category induction. There were a lot of studies about neural activities of uncertainty in the areas of decision-making, mainly using guessing task, gambling task, forecasting task based on the probability.Although predecessors have conducted many studies in detail on the category learning, there were some problems. Due to different criteria for categorization, variety of experimental paradigms, different categorization strategies on different tasks, and different purpose for studies, there was no unified conclusion of neural activity about category learning in many studies. So investigation of the function of different brain regions in representation, storage and recall and levels of categorization was a major challenge in the future. In addition, studies on category learning focused on patient population, it might be not comprehensive. The quantity of patients was less in most of studies, and there were differences between individual such as age, personality, and education differences, etc. These individual differences resulted in different findings obtained by researchers, which was difficult to be described as a result of the impact of the independent variables; these results might be due to the subject factor, or interaction effect between independent variables and subject factor. So far it was still lacking for the investigation of the mental process and neural activity of category induction. Bigman (2004) and Chen etc.(2007,2008) made a full analysis for mental process of category induction on the basis of previous studies, and explored the neural mechanisms of category learning using high temporal resolution of ERP, but the lack of follow-up studies. These studies did not adopt the high spatial resolution of fMRI technology to explore neural mechanisms of category induction. In addition, another important sub-process of category induction, i.e. the uncertainty, has not been investigated. Hence, the issue about neural network of category induction was needed to be explored urgently.For the above problems, this paper used two studies to explore neural meachnism of category induction.First, this study refers to the invariance of concept learning of Vigo (2013). Invariance means when one dimension is ignored or inhibited, the elements of one category are extracted or filter out. The essence of invariance is that, the more the number of invariance is, the easier category learning is; the less the number of invariance is, the harder category learning is. In order to investigate neural mechanisms of invariant detection in category induction, based on previous research, especially the complex cognitive segment design of Chen, we explored neural activities of the invariance, activation loop among brain regions and its laws in category induction, using a step-wise methods by fMRI technique, combining cognitive neuroscience with behavioral research. The reason for using the step-wise methods, the considerations is as follows:Firstly, the paradigm has been used before. Secondly, category structure is clear and cognitive processes are simple in this task, it is good for exploring neural activity of searching for commonalities in simple category learning task. Therefore, we designed three experiments to try to explore neural mechanism of invariant detection in category induction. The results found:(1) With the process of invariance detection of category induction, it is active in left middle frontal gyrus. The PPI analysis found that interaction effect increase between parietal and left middle frontal gyrus in the final stage of category induction. These results revealed an important role of left prefrontal-parietal network in the invariant detection of category induction, and the initial demonstration that the left middle frontal gyrus may be the core area of invariance detection in category induction. (2) The exclusion of expectation and addition the degree of complex of experimental task, we found invariance detection activates the left prefrontal (BAs 6,9, and 44), bilateral parietal (BA 7) and striatum. The ROI analysis and PPI analysis found that the prefrontal-parietal-striatum network was activated strongly when more null hypothesis is rejected. With the increase of the amount of detecting invariance and inhibiting variance, the brain functional connectivity of prefrontal with posterior regions became stronger, especially with the parietal lobe (inferior parietal, precuneus), caudate nucleus, putamen, and temporal lobe. More importantly, with the same amount of inhibiting features, the difference between extracting one invariant feature and extracting two invariant ones was reflected by the activation of left dorsolateral prefrontal cortex (DLPFC). The results further demonstrated the process of invariance detection was closely related to DLPFC. (3) As much as possible to control and reduce the effect of working memory, we also found activation of left superior frontal gyrus (BA 6) and left middle/inferior frontal gyrus (BA 8,9) in the invariance detection.The results from three experiments revealed the intrinsic relationships among prefrontal-parietal, striatum, invariant detection and working memory. The prefrontal-parietal network was mainly involved in processing of invariance induction. More importantly, the left DLPFC (BA 6,8,9) might be the core region of invariance detection. The activation of striatum was closely related to working memory. When the stimuli representation was stored in working memory, and induction process made for the next phase by using representation information, the striatum played an important role.Second, we explored the neural mechanism of categorization uncertainty in different category induction conditions formed by changing the degree of invariance among simple stimuli. In other words, the specific brain activation and activity patterns among regional networks were mainly explored about uncertainty of category induction by using fMRI technology. We investigated the uncertainty in Experiment 4 and 5. The results found:(1) Within the greater the uncertainty of inductive conclusion, the more intense the left inferior parietal lobe activated. But in the more certainty of category induction, there were stronger activitions in left ACC. The conclusion might indicate that the left inferior parietal lobe was closely related to uncertainty in category induction. But the preliminary conclusion was obtained from simple category induction task (Experiment 4). The process of category induction and categorization was made step by step. The participants made a conclusion under the requirement of working memory. The activation region of cognitive process of working memory may overlap with the categorization uncertainty. In addition, the set of experimental condition was not relatively complex, and lack of comprehensive investigation for the neural mechanism of categorization uncertainty. (2) In control of working memory and the addition of category induction condition, we further reached a conclusion in the Experiment 5 that uncertainty mainly activated left inferior parietal lobe, superior parietal lobe, left middle frontal gyrus, left inferior frontal gyrus and bilateral insula etc. With the increase of uncertainty, these regions were more active, especially the left inferior and superior parietal lobe. We initially made a conclusion that the left inferior and superior parietal lobe was the core activation region of categorization uncertainty. The uncertainty was regulated by prefrontal cortex "executive control system", and response selection decisions induce uncertainty choice state, making a right response to stimulus also lead to the activation of insula without extrinsic rewards driven.Based on the results revealed in this paper and previous research, we proposed that neural networks about invariance detection of category induction and categorization uncertainty. Category induction was relevant with frontal-parietal-striatum network. The dorsolateral prefrontal cortex was mainly involved in the process of invariance detection. When more working memory was required in tasks, the striatum kept information in WM, and made inductive process by using intrinsic representation. In the stage of categorization, parietal region, specifically left inferior and superior parietal lobe, played a very crucial role in the process of categorization uncertainty.