Never-Ending Learning for Explainable Brain Computing

Exploring the nature of human intelligence and behavior is a longstanding pursuit in cognitive neuroscience, driven by the accumulation of knowledge, information, and data across various studies. However, achieving a unified and transparent interpretation of findings presents formidable challenges. In response, an explainable brain computing framework is proposed that employs the never-ending learning paradigm, integrating evidence combination and fusion computing within a Knowledge-Information-Data (KID) architecture. The framework supports continuous brain cognition investigation, utilizing joint knowledge-driven forward inference and data-driven reverse inference, bolstered by the pre-trained language modeling techniques and the human-in-the-loop mechanisms. In particular, it incorporates internal evidence learning through multi-task functional neuroimaging analyses and external evidence learning via topic modeling of published neuroimaging studies, all of which involve human interactions at different stages. Based on two case studies, the intricate uncertainty surrounding brain localization in human reasoning is revealed. The present study also highlights the potential of systematization to advance explainable brain computing, offering a finer-grained understanding of brain activity patterns related to human intelligence.

No effects of the theta-frequency transcranial electrical stimulation for recall, attention control, and relation integration in working memory

Introduction

Recent studies have suggested that transcranial alternating current stimulation (tACS), and especially the theta-frequency tACS, can improve human performance on working memory tasks. However, evidence to date is mixed. Moreover, the two WM tasks applied most frequently, namely the n-back and change-detection tasks, might not constitute canonical measures of WM capacity.

Method

In a relatively large sample of young healthy participants (N = 62), we administered a more canonical WM task that required stimuli recall, as well as we applied two WM tasks tapping into other key WM functions: attention control (the antisaccade task) and relational integration (the graph mapping task). The participants performed these three tasks three times: during the left frontal 5.5-Hz and the left parietal 5.5-Hz tACS session as well as during the sham session, with a random order of sessions. Attentional vigilance and subjective experience were monitored.

Results

For each task administered, we observed significant gains in accuracy neither for the frontal tACS session nor for the parietal tACS session, as compared to the sham session. By contrast, the scores on each task positively inter-correlated across the three sessions.

Discussion

The results suggest that canonical measures of WM capacity are strongly stable in time and hardly affected by theta-frequency tACS. Either the tACS effects observed in the n-back and change detection tasks do not generalize onto other WM tasks, or the tACS method has limited effectiveness with regard to WM, and might require further methodological advancements.

Electrophysiological connectivity of logical deduction: Early cortical MEG study

Complex human reasoning involves minimal abilities to extract conclusions implied in the available information. These abilities are considered "deductive" because they exemplify certain abstract relations among propositions or probabilities called deductive arguments. However, the electrophysiological dynamics which supports such complex cognitive processes has not been addressed yet. In this work we consider typically deductive logico-probabilistically valid inferences and aim to verify or refute their electrophysiological functional connectivity differences from invalid inferences with the same content (same relational variables, same stimuli, same relevant and salient features). We recorded the brain electrophysiological activity of 20 participants (age = 20.35 ± 3.23) by means of an MEG system during two consecutive reasoning tasks: a search task (invalid condition) without any specific deductive rules to follow, and a logically valid deductive task (valid condition) with explicit deductive rules as instructions. We calculated the functional connectivity (FC) for each condition and conducted a seed-based analysis in a set of cortical regions of interest. Finally, we used a cluster-based permutation test to compare the differences between logically valid and invalid conditions in terms of FC. As a first novel result we found higher FC for valid condition in beta band between regions of interest and left prefrontal, temporal, parietal, and cingulate structures. FC analysis allows a second novel result which is the definition of a propositional network with operculo-cingular, parietal and medial nodes, specifically including disputed medial deductive "core" areas. The experiment discloses measurable cortical processes which do not depend on content but on truth-functional propositional operators. These experimental novelties may contribute to understand the cortical bases of deductive processes.

Probabilistic and Deductive Reasoning in the Human Brain

Reasoning is a process of inference from given premises to new conclusions. Deductive reasoning is truth-preserving and conclusions can only be either true or false. Probabilistic reasoning is based on degrees of belief and conclusions can be more or less likely. While deductive reasoning requires people to focus on the logical structure of the inference and ignore its content, probabilistic reasoning requires the retrieval of prior knowledge from memory. Recently, however, some researchers have denied that deductive reasoning is a faculty of the human mind. What looks like deductive inference might actually also be probabilistic inference, only with extreme probabilities. We tested this assumption in an fMRI experiment with two groups of participants: one group was instructed to reason deductively, the other received probabilistic instructions. They could freely choose between a binary and a graded response to each problem. The conditional probability and the logical validity of the inferences were systematically varied. Results show that prior knowledge was only used in the probabilistic reasoning group. These participants gave graded responses more often than those in the deductive reasoning group and their reasoning was accompanied by activations in the hippocampus. Participants in the deductive group mostly gave binary responses and their reasoning was accompanied by activations in the anterior cingulate cortex, inferior frontal cortex, and parietal regions. These findings show that (1) deductive and probabilistic reasoning rely on different neurocognitive processes, (2) people can suppress their prior knowledge to reason deductively, and (3) not all inferences can be reduced to probabilistic reasoning.

Neural correlates of transitive inference: An SDM meta-analysis on 32 fMRI studies

Transitive inference (TI) is a critical capacity involving the integration of relevant information into prior knowledge structure for drawing novel inferences on unobserved relationships. To date, the neural correlates of TI remain unclear due to the small sample size and heterogeneity of various experimental tasks from individual studies. Here, the meta-analysis on 32 fMRI studies was performed to detect brain activation patterns of TI and its three paradigms (spatial inference, hierarchical inference, and associative inference). We found the hippocampus, prefrontal cortex (PFC), putamen, posterior parietal cortex (PPC), retrosplenial cortex (RSC), supplementary motor area (SMA), precentral gyrus (PreCG), and median cingulate cortex (MCC) were engaged in TI. Specifically, the RSC was implicated in the associative inference, whereas PPC, SMA, PreCG, and MCC were implicated in the hierarchical inference. In addition, the hierarchical inference and associative inference both evoked activation in the hippocampus, medial PFC, and PCC. Although the meta-analysis on spatial inference did not generate a reliable result due to insufficient amount of investigations, the present work still offers a new insight for better understanding the neural basis underlying TI.

Common and distinct neurofunctional representations of core and social disgust in the brain: Coordinate-based and network meta-analyses

Disgust represents a multifaceted defensive-avoidance response. On the behavioral level, the response includes withdrawal and a disgust-specific facial expression. While both serve the avoidance of pathogens, the latter additionally transmits social-communicative information. Given that common and distinct brain representation of the primary defensive-avoidance response (core disgust) and encoding of the social-communicative signal (social disgust) remain debated, we employed neuroimaging meta-analyses to (1) determine brain systems generally engaged in disgust processing, and (2) segregate common and distinct brain systems for core and social disgust. Disgust processing, in general, engaged a bilateral network encompassing the insula, amygdala, occipital and prefrontal regions. Core disgust evoked stronger reactivity in left-lateralized threat detection and defensive response network including amygdala, occipital and frontal regions, while social disgust engaged a right-lateralized superior temporal-frontal network engaged in social cognition. Anterior insula, inferior frontal and fusiform regions were commonly engaged during core and social disgust, suggesting a shared neurofunctional basis. We demonstrate a common and distinct neural basis of primary disgust responses and encoding of associated social-communicative signals.

A tale of two gradients: differences between the left and right hemispheres predict semantic cognition

Decomposition of whole-brain functional connectivity patterns reveals a principal gradient that captures the separation of sensorimotor cortex from heteromodal regions in the default mode network (DMN). Functional homotopy is strongest in sensorimotor areas, and weakest in heteromodal cortices, suggesting there may be differences between the left and right hemispheres (LH/RH) in the principal gradient, especially towards its apex. This study characterised hemispheric differences in the position of large-scale cortical networks along the principal gradient, and their functional significance. We collected resting-state fMRI and semantic, working memory and non-verbal reasoning performance in 175 + healthy volunteers. We then extracted the principal gradient of connectivity for each participant, tested which networks showed significant hemispheric differences on the gradient, and regressed participants’ behavioural efficiency in tasks outside the scanner against interhemispheric gradient differences for each network. LH showed a higher overall principal gradient value, consistent with its role in heteromodal semantic cognition. One frontotemporal control subnetwork was linked to individual differences in semantic cognition: when it was nearer heteromodal DMN on the principal gradient in LH, participants showed more efficient semantic retrieval—and this network also showed a strong hemispheric difference in response to semantic demands but not working memory load in a separate study. In contrast, when a dorsal attention subnetwork was closer to the heteromodal end of the principal gradient in RH, participants showed better visual reasoning. Lateralization of function may reflect differences in connectivity between control and heteromodal regions in LH, and attention and visual regions in RH.

A Cortical Surface-Based Meta-Analysis of Human Reasoning

Recent advances in neuroimaging have augmented numerous findings in the human reasoning process but have yielded varying results. One possibility for this inconsistency is that reasoning is such an intricate cognitive process, involving attention, memory, executive functions, symbolic processing, and fluid intelligence, whereby various brain regions are inevitably implicated in orchestrating the process. Therefore, researchers have used meta-analyses for a better understanding of neural mechanisms of reasoning. However, previous meta-analysis techniques include weaknesses such as an inadequate representation of the cortical surface’s highly folded geometry. Accordingly, we developed a new meta-analysis method called Bayesian meta-analysis of the cortical surface (BMACS). BMACS offers a fast, accurate, and accessible inference of the spatial patterns of cognitive processes from peak brain activations across studies by applying spatial point processes to the cortical surface. Using BMACS, we found that the common pattern of activations from inductive and deductive reasoning was colocalized with the multiple-demand system, indicating that reasoning is a high-level convergence of complex cognitive processes. We hope surface-based meta-analysis will be facilitated by BMACS, bringing more profound knowledge of various cognitive processes.

The Development of Relational Reasoning in South Korean Elementary and Middle-School Students: A Cross-Sectional Investigation

Relational reasoning is a higher-order executive function that involves the ability to perceive meaningful patterns within a body of seemingly unrelated information. In this study, the ability of 749 fourth (M_age = 10), sixth (M_age = 12), eighth (M_age = 14), and tenth graders (M_age = 16) to identify meaningful relational patterns was investigated. This general cognitive ability was assessed by means of the Test of Relational Reasoning-Junior (TORRjr), a 32-item measure organized into four 8-item scales that assess analogical, anomalous, antinomous, and antithetical reasoning. Students’ performance on the TORRjr was analyzed using confirmatory factor analysis, measurement invariance test, and non-parametric median-based analyses. The confirmatory factor analysis supported that the higher-order factor model was the best fit for the TORRjr data for the Korean students. The measurement was determined to be invariant by gender but variant across grade levels. The non-parametric analysis resulted in an asymptotic (a constant increasing up to grade 6 and then a level off witnessed from grades 8 to 10) development pattern in overall relational reasoning across the grades. In comparison to analogy and anomaly, antinomy and antithesis scores were more fully developed by grade 8 and that level of performance was maintained at grade 10. The TORRjr appeared to be a viable measure for the Korean samples up to approximately 15 years of age. The significance of these findings for research and instructional practice are discussed.

The neural bases of argumentative reasoning

Most reasoning tasks used in behavioral and neuroimaging studies are abstract, triggering slow, effortful processes. By contrast, most of everyday life reasoning is fast and effortless, as when we exchange arguments in conversation. Recent behavioral studies have shown that reasoning tasks with the same underlying logic can be solved much more easily if they are embedded in an argumentative context. In the present article, we study the neural bases of this type of everyday, argumentative reasoning. Such reasoning is both a social and a metarepresentational process, suggesting it should share some mechanisms, and thus some neural bases, with other social, metarepresentational process such as pragmatics, metacognition, or theory of mind. To isolate the neural bases of argumentative reasoning, we measured fMRI activity of participants who read the same statement presented either as the conclusion of an argument, or as an assertion. We found that conclusions of arguments, compared to assertions, were associated with greater activity in a region of the medial prefrontal cortex that was identified in quantitative meta-analyses of studies on theory of mind. This study shows that it is possible to use more ecologically valid tasks to study the neural bases of reasoning, and that using such tasks might point to different neural bases than those observed with the more abstract and artificial tasks typically used in the neuroscience of reasoning. Specifically, we speculate that reasoning in an argumentative context might rely on mechanisms supporting metarepresentational processes in the medial prefrontal cortex.

Relational Integration in the Human Brain: A Review and Synthesis

Relational integration is required when multiple explicit representations of relations between entities must be jointly considered to make inferences. We provide an overview of the neural substrate of relational integration in humans and the processes that support it, focusing on work on analogical and deductive reasoning. In addition to neural evidence, we consider behavioral and computational work that has informed neural investigations of the representations of individual relations and of relational integration. In very general terms, evidence from neuroimaging, neuropsychological, and neuromodulatory studies points to a small set of regions (generally left lateralized) that appear to constitute key substrates for component processes of relational integration. These include posterior parietal cortex, implicated in the representation of first-order relations (e.g., A:B); rostrolateral pFC, apparently central in integrating first-order relations so as to generate and/or evaluate higher-order relations (e.g., A:B::C:D); dorsolateral pFC, involved in maintaining relations in working memory; and ventrolateral pFC, implicated in interference control (e.g., inhibiting salient information that competes with relevant relations). Recent work has begun to link computational models of relational representation and reasoning with patterns of neural activity within these brain areas.

Neural representations of transitive relations predict current and future math calculation skills in children

A large body of evidence suggests that math learning in children is built upon innate mechanisms for representing numerical quantities in the intraparietal sulcus (IPS). Learning math, however, is about more than processing quantitative information. It is also about understanding relations between quantities and making inferences based on these relations. Consistent with this idea, recent behavioral studies suggest that the ability to process transitive relations (A > B, B > C, therefore A > C) may contribute to math skills in children. Here we used fMRI coupled with a longitudinal design to determine whether the neural processing of transitive relations in children could predict their current and future math skills. At baseline (T₁), children (n = 31) processed transitive relations in an MRI scanner. Math skills were measured at T₁ and again 1.5 years later (T₂). Using a machine learning approach with cross-validation, we found that activity associated with the representation of transitive relations in the IPS predicted math calculation skills at both T₁ and T₂. Our study highlights the potential of neurobiological measures of transitive reasoning for forecasting math skills in children, providing additional evidence for a link between this type of reasoning and math learning.

The Neurocognitive Correlates of Human Reasoning: A Meta-analysis of Conditional and Syllogistic Inferences

Inferring knowledge is a core aspect of human cognition. We can form complex sentences connecting different pieces of information, such as in conditional statements like "if someone drinks alcohol, then they must be older than 18." These are relevant for causal reasoning about our environment and allow us to think about hypothetical scenarios. Another central aspect to forming complex statements is to quantify about sets, such as in "some apples are green." Reasoning in terms of the ability to form these statements is not yet fully understood, despite being an active field of interdisciplinary research. On a theoretical level, several conceptual frameworks have been proposed, predicting diverging brain activation patterns during the reasoning process. We present a meta-analysis comprising the results of 32 neuroimaging experiments about reasoning, which we subdivided by their structure, content, and requirement for world knowledge. In conditional tasks, we identified activation in the left middle and rostrolateral pFC and parietal regions, whereas syllogistic tasks elicit activation in Broca's complex, including the BG. Concerning the content differentiation, abstract tasks exhibit activation in the left inferior and rostrolateral pFC and inferior parietal regions, whereas content tasks are in the left superior pFC and parieto-occipital regions. The findings clarify the neurocognitive mechanisms of reasoning and exhibit clear distinctions between the task's type and content. Overall, we found that the activation differences clarify inconsistent results from accumulated data and serve as useful scaffolding differentiations for theory-driven interpretations of the neuroscientific correlates of human reasoning.