Relational Reasoning Tasks Research Articles

Skin lesion classification based on hybrid self‐supervised pretext task

AbstractThe combination of observation of skin lesion and digital image technology contributes to the diagnosis and treatment of skin diseases. To solve the problems of large variation of target size and shape in skin disease images, small difference between disease images and normal images, and difficulty of label acquisition, we propose a classification algorithm for skin lesion based on hybrid self‐supervised pretext tasks. We combine the discriminative self‐supervised relational reasoning task with the generative self‐supervised mutual information maximization task and design the loss function. Experiments show that the generative‐discriminative self‐supervised co‐training algorithm can identify images with small differences and learn discriminative features and achieve a classification accuracy of 82.6% on the DermaMNIST dataset.

Experimental investigation of training schedule on home-based working memory training in healthy older adults.

The efficacy of working memory training (WMT) for cognitive enhancement in healthy older adults has been extensively investigated. Typically, WMT results in improved performance on the training task, but limited or no transfer of improvement to other cognitive tasks. Accordingly, there is a need to identify optimal intervention parameters to maximize training and transfer task effects of WMT. The current study aimed to investigate the effect of training schedule on training and transfer task performance of WMT in healthy older adults. A secondary aim was to examine the feasibility of participants performing the intervention online at home, unsupervised, and using their personal devices. Participants (N = 71; mean age: 66 years) completed sixteen WMT or active-control sessions over eight (distributed) or four (intensive) weeks. Adaptive verbal and spatial n-back tasks were used as the WMT tasks. We tested near transfer effects to a digit-span task and far transfer effects to an abstract relational reasoning task. Participants successfully performed the cognitively demanding intervention using their own devices, online at home, and with minimal contact with the researcher. We observed a significant improvement in WMT task performance in the WMT group relative to active-controls, but no evidence of near or far transfer. Similar training effects were observed irrespective of the intensity of the training schedule. Our results suggest that comparable benefits could be observed when using less intensive schedules that may be more easily accommodated into everyday life.

Generating distant analogies facilitates relational integration: Intermediary role of relational mindset and cognitive load.

Relational integration is essential for learning, working, and living, as we must encode enormous volumes of information and extract their relations to construct knowledge about the environment. Recent research hints that generating distant analogies can temporarily facilitate learners’ state-based relational integration. This study aimed to investigate the internal mechanism underlying the facilitation effect and preliminarily confirm its application in education. First, we adopted the classical n-term premise integration task (Experiment 1a) and the Latin Square Task (Experiment 1b) to explore the robustness of the facilitation effect. Then we employed an emerging multidimensional relational reasoning task to further explore the internal mechanism underlying this facilitation effect (Experiment 2). Finally, we verified the practical role of the facilitation effect in learning the interaction concept in statistics (Experiment 3). The results showed that generating distant analogies did facilitate students’ relational integration performance, both in classical cognitive tasks and in a practical learning task, and a relational mindset and cognitive load play an intermediary role in the facilitation, supporting the cognitive load theory. The results suggest that generating distant analogies can be a useful warm-up activity to assist educators in promoting students’ relational integration.

Fluid intelligence and its neural correlates in early onset mild cognitive impairment

AbstractBackgroundFluid intelligence refers to the ability of abstract reasoning, problem‐solving and other mental activities considered to be independent from learned knowledges and experience. It has been demonstrated that fluid intelligence is compromised in Mild Cognitive Impairment (MCI), but it also declines with ageing. We tested if fluid intelligence is also affected when MCI occurs before 65 years of age (i.e., in early‐onset MCI) and studied its specific functional and structural brain correlates independently from the effect of ageing.MethodEighty‐five subjects with early‐onset MCI (cognitive symptoms onset < 65 years) underwent extended neuropsychological assessment including Raven’s Coloured Progressive Matrices (RCPM), a culture‐free non‐verbal relational reasoning task that is considered a proxy of fluid intelligence. They also underwent multimodal Magnetic Resonance Imaging (MRI) protocol on a 3T scanner including high‐resolution T1 and sequences acquired at rest for functional MRI (fMRI). Performance at RCPM was related to regional grey matter volume using Voxel Based Morphometry (VBM) and to functional connectivity of resting‐state networks (RSNs) known to be related to cognition using independent component analysis.ResultsThe median RCPM score was 27 (range 8‐36, with 36 being the maximum obtainable score). Among all 85 patients, only 11.8% (N=10) performed below the age and education‐corrected cut‐off. These patients also had significantly lower scores in almost all the remaining neuropsychological tests, suggesting that they were in a more severe stage of MCI (i.e., close to conversion). VBM results showed that, across all patients, decreased performance on RCPM was associated with focal atrophy in right fronto‐temporal and bilateral parietal areas (Figure 1). FMRI acquired at rest showed that better performance on RCPM was associated with lower functional connectivity in areas of the anterior DMN and with higher functional connectivity in areas of the left fronto‐parietal RSN (also indicated as the working memory RSN).ConclusionIn patients with early‐onset MCI fluid intelligence is relatively preserved, associated with lower atrophy in right fronto‐temporal and bilateral parietal regions, and with decreased functional connectivity in the anterior DMN and increased in the left fronto‐parietal network.

Evaluating the progress of deep learning for visual relational concepts.

Convolutional neural networks have become the state-of-the-art method for image classification in the last 10 years. Despite the fact that they achieve superhuman classification accuracy on many popular datasets, they often perform much worse on more abstract image classification tasks. We will show that these difficult tasks are linked to relational concepts from cognitive psychology and that despite progress over the last few years, such relational reasoning tasks still remain difficult for current neural network architectures. We will review deep learning research that is linked to relational concept learning, even if it was not originally presented from this angle. Reviewing the current literature, we will argue that some form of attention will be an important component of future systems to solve relational tasks. In addition, we will point out the shortcomings of currently used datasets, and we will recommend steps to make future datasets more relevant for testing systems on relational reasoning.

Distributed associative memory network with memory refreshing loss

Despite recent progress in memory augmented neural network (MANN) research, associative memory networks with a single external memory still show limited performance on complex relational reasoning tasks. Especially the content-based addressable memory networks often fail to encode input data into rich enough representation for relational reasoning and this limits the relation modeling performance of MANN for long temporal sequence data. To address these problems, here we introduce a novel Distributed Associative Memory architecture (DAM) with Memory Refreshing Loss (MRL) which enhances the relation reasoning performance of MANN. Inspired by how the human brain works, our framework encodes data with distributed representation across multiple memory blocks and repeatedly refreshes the contents for enhanced memorization similar to the rehearsal process of the brain. For this procedure, we replace a single external memory with a set of multiple smaller associative memory blocks and update these sub-memory blocks simultaneously and independently for the distributed representation of input data. Moreover, we propose MRL which assists a task’s target objective while learning relational information existing in data. MRL enables MANN to reinforce an association between input data and task objective by reproducing stochastically sampled input data from stored memory contents. With this procedure, MANN further enriches the stored representations with relational information. In experiments, we apply our approaches to Differential Neural Computer (DNC), which is one of the representative content-based addressing memory models and achieves the state-of-the-art performance on both memorization and relational reasoning tasks.

What Makes Mental Modeling Difficult? Normative Data for the Multidimensional Relational Reasoning Task.

Relational reasoning is a complex form of human cognition involving the evaluation of relations between mental representations of information. Prior studies have modified stimulus properties of relational reasoning problems and examined differences in difficulty between different problem types. While subsets of these stimulus properties have been addressed in separate studies, there has not been a comprehensive study, to our knowledge, which investigates all of these properties in the same set of stimuli. This investigative gap has resulted in different findings across studies which vary in task design, making it challenging to determine what stimulus properties make relational reasoning—and the putative formation of mental models underlying reasoning—difficult. In this article, we present the Multidimensional Relational Reasoning Task (MRRT), a task which systematically varied an array of stimulus properties within a single set of relational reasoning problems. Using a mixed-effects framework, we demonstrate that reasoning problems containing a greater number of the premises as well as multidimensional relations led to greater task difficulty. The MRRT has been made publicly available for use in future research, along with normative data regarding the relative difficulty of each problem.

Assessing the Effects of a Relational Training Intervention on Fluid Intelligence Among a Sample of Socially Disadvantaged Children in Bangladesh

A growing consensus is that performances on standardized intelligence tests can be positively influenced by interventions that focus on improving relational reasoning. One such intervention, known as SMART (Strengthening Mental Abilities with Relational Training), consists of presenting participants with increasingly complex exemplars of relational reasoning tasks involving premises associated along same-as, opposite-of, more-than, and less-than relations. Following multiple training and testing sessions involving these relations, increased fluency in SMART is related to enhanced performances on tests of intellectual ability (Cassidy et al., 2016). The current study expands upon previous investigations on SMART in two ways. First, we explored whether the amount of training undertaken can predict changes in intelligence test performances. Second, we assessed whether SMART training could be effective for a non-English speaking, socioeconomically disadvantaged cohort. Changes in intelligence were assessed via administrations of the standard Raven’s Progressive Matrices before and after SMART training. Our results show that the stage of SMART training completed is positively related to changes in Ravens’ performances and that such training can be effective for non-English cohorts.

Discriminating relational and perceptual judgments: Evidence from human toddlers

The ability to represent same-different relations is an important condition for abstract thought. However, there is mixed evidence for when this ability develops, both ontogenetically and phylogenetically. Apparent success in relational reasoning may be evidence for genuine conceptual understanding or may be the result of low-level, perceptual strategies. We introduce a method to discriminate these possibilities by pitting two conditions that are perceptually matched but conceptually different: in a “fused” condition, same and different objects are joined, creating single objects that have the same perceptual features as the two object pairs in the “relational” condition. However, the “fused” objects do not provide evidence for the relation ‘same.’ Using this method with human toddlers in a causal relational reasoning task provides evidence for genuine conceptual understanding. This novel technique offers a simple manipulation that may be applied to a variety of existing match-to-sample procedures used to assess same-different reasoning to include in future research with non-human animals across species, as well as human infants.

Audience effects on the neural correlates of relational reasoning in adolescence

Adolescents are particularly sensitive to peer influence. This may partly be due to an increased salience of peers during adolescence. We investigated the effect of being observed by a peer on a cognitively challenging task, relational reasoning, which requires the evaluation and integration of multiple mental representations. Relational reasoning tasks engage a fronto-parietal network including the inferior parietal cortex, pre-supplementary motor area, dorsolateral and rostrolateral prefrontal cortices. Using functional magnetic resonance imaging (fMRI), peer audience effects on activation in this fronto-parietal network were compared in a group of 19 female mid-adolescents (aged 14–16 years) and 14 female adults (aged 23–28 years). Adolescent and adult relational reasoning accuracy was influenced by a peer audience as a function of task difficulty: the presence of a peer audience led to decreased accuracy in the complex, relational integration condition in both groups of participants. The fMRI results demonstrated that a peer audience differentially modulated activation in regions of the fronto-parietal network in adolescents and adults. Activation was increased in adolescents in the presence of a peer audience, while this was not the case in adults.

General and specialized brain correlates for analogical reasoning: A meta-analysis of functional imaging studies.

Reasoning by analogy allows us to link distinct domains of knowledge and to transfer solutions from one domain to another. Analogical reasoning has been studied using various tasks that have generally required the consideration of the relationships between objects and their integration to infer an analogy schema. However, these tasks varied in terms of the level and the nature of the relationships to consider (e.g., semantic, visuospatial). The aim of this study was to identify the cerebral network involved in analogical reasoning and its specialization based on the domains of information and task specificity. We conducted a coordinate-based meta-analysis of 27 experiments that used analogical reasoning tasks. The left rostrolateral prefrontal cortex was one of the regions most consistently activated across the studies. A comparison between semantic and visuospatial analogy tasks showed both domain-oriented regions in the inferior and middle frontal gyri and a domain-general region, the left rostrolateral prefrontal cortex, which was specialized for analogy tasks. A comparison of visuospatial analogy to matrix problem tasks revealed that these two relational reasoning tasks engage, at least in part, distinct right and left cerebral networks, particularly separate areas within the left rostrolateral prefrontal cortex. These findings highlight several cognitive and cerebral differences between relational reasoning tasks that can allow us to make predictions about the respective roles of distinct brain regions or networks. These results also provide new, testable anatomical hypotheses about reasoning disorders that are induced by brain damage. Hum Brain Mapp 37:1953-1969, 2016. © 2016 Wiley Periodicals, Inc.

Sensitivity to Relational Similarity and Object Similarity in Apes and Children

Relational reasoning is a hallmark of sophisticated cognition in humans. Does it exist in other primates? Despite some affirmative answers, there appears to be a wide gap in relational ability between humans and other primates--even other apes. Here, we test one possible explanation for this gap, motivated by developmental research showing that young humans often fail at relational reasoning tasks because they focus on objects instead of relations. When asked, "duck:duckling is like tiger:?," preschool children choose another duckling (object match) rather than a cub. If other apes share this focus on concrete objects, it could undermine their relational reasoning in similar ways. To test this, we compared great apes and 3-year-old humans' relational reasoning on the same spatial mapping task, with and without competing object matches. Without competing object matches, both children and Pan species (chimpanzees and bonobos) spontaneously used relational similarity, albeit children more so. But when object matches were present, only children responded strongly to them. We conclude that the relational gap is not due to great apes' preference for concrete objects. In fact, young humans show greater object focus than nonhuman apes.

Developmental changes in effective connectivity associated with relational reasoning.

Rostrolateral prefrontal cortex (RLPFC) is part of a frontoparietal network of regions involved in relational reasoning, the mental process of working with relationships between multiple mental representations. RLPFC has shown functional and structural changes with age, with increasing specificity of left RLPFC activation for relational integration during development. Here, we used dynamic causal modeling (DCM) to investigate changes in effective connectivity during a relational reasoning task through the transition from adolescence into adulthood. We examined fMRI data of 37 healthy female participants (11–30 years old) performing a relational reasoning paradigm. Comparing relational integration to the manipulation of single relations revealed activation in five regions: the RLPFC, anterior insula, dorsolateral PFC, inferior parietal lobe, and medial superior frontal gyrus. We used a new exhaustive search approach and identified a full DCM model, which included all reciprocal connections between the five clusters in the left hemisphere, as the optimal model. In line with previous resting state fMRI results, we showed distinct developmental effects on the strength of long-range frontoparietal versus frontoinsular short-range fixed connections. The modulatory connections associated with relational integration increased with age. Gray matter volume in left RLPFC, which decreased with age, partly accounted for changes in fixed PFC connectivity. Finally, improvements in relational integration performance were associated with greater modulatory and weaker fixed PFC connectivity. This pattern provides further evidence of increasing specificity of left PFC function for relational integration compared to the manipulation of single relations, and demonstrates an association between effective connectivity and performance during development.

Development of relational reasoning during adolescence

Non-linear changes in behaviour and in brain activity during adolescent development have been reported in a variety of cognitive tasks. These developmental changes are often interpreted as being a consequence of changes in brain structure, including non-linear changes in grey matter volumes, which occur during adolescence. However, very few studies have attempted to combine behavioural, functional and structural data. This multi-method approach is the one we took in the current study, which was designed to investigate developmental changes in behaviour and brain activity during relational reasoning, the simultaneous integration of multiple relations. We used a relational reasoning task known to recruit rostrolateral prefrontal cortex (RLPFC), a region that undergoes substantial structural changes during adolescence. The task was administered to female participants in a behavioural (N = 178, 7-27 years) and an fMRI study (N = 37, 11-30 years). Non-linear changes in accuracy were observed, with poorer performance during mid-adolescence. fMRI and VBM results revealed a complex picture of linear and possibly non-linear changes with age. Performance and structural changes partly accounted for changes with age in RLPFC and medial superior frontal gyrus activity but not for a decrease in activation in the anterior insula/frontal operculum between mid-adolescence and adulthood. These functional changes might instead reflect the maturation of neurocognitive strategies.

Neurocognitive development of relational reasoning

Relational reasoning is an essential component of fluid intelligence, and is known to have a protracted developmental trajectory. To date, little is known about the neural changes that underlie improvements in reasoning ability over development. In this event-related functional magnetic resonance imaging (fMRI) study, children aged 8-12 and adults aged 18-25 performed a relational reasoning task adapted from Raven's Progressive Matrices. The task included three levels of relational reasoning demands: REL-0, REL-1, and REL-2. Children exhibited disproportionately lower accuracy than adults on trials that required integration of two relations (REL-2). Like adults, children engaged lateral prefrontal cortex (PFC) and parietal cortex during task performance; however, they exhibited different time courses and activation profiles, providing insight into their approach to the problems. As in prior studies, adults exhibited increased rostrolateral PFC (RLPFC) activation when relational integration was required (REL-2 > REL-1, REL-0). Children also engaged RLPFC most strongly for REL-2 problems at early stages of processing, but this differential activation relative to REL-1 trials was not sustained throughout the trial. These results suggest that the children recruited RLPFC while processing relations, but failed to use it to integrate across two relations. Relational integration is critical for solving a variety of problems, and for appreciating analogies; the current findings suggest that developmental improvements in this function rely on changes in the profile of engagement of RLPFC, as well as dorsolateral PFC and parietal cortex.

Adult Age Differences in Function Concept Learning

ABSTRACT Function concept learning and knowledge use was explored across adulthood. During training older and younger adults predicted an amount of physiological arousal produced as a negative and positive function of a chemical substance. Knowledge use was evaluated with two transfer conditions requiring a switch between contextual contingencies: a relationship inversion, predicting the chemical amount given the physiological arousal, and a change from graphic based to text based stimuli. Older adults were impaired in applying the negative slope concept. However, there was no relative deficit in switching between the negative and positive function slopes or inverting the learned relationship. Our results suggest that age-related differences in relational reasoning tasks vary not only with processing efficiency, but also task related conceptual knowledge.

The role of relational reasoning in children's addition concepts

The study addresses the relational reasoning of different‐aged children and how addition reasoning is related to problem‐solving skills within addition and to reasoning skills outside addition. Ninety‐two 5‐ to 8‐year‐olds were asked to solve a series of conceptually related and unrelated addition problems, and the speed and accuracy of all self‐reported strategies were used to monitor their addition performance. Children were also given a series of general relational reasoning tasks to assess their ability to solve problems based on thematic, causal and visual relations. The results revealed that, while children were able to reason about commutativity relations, recognition of relations based on additive composition was rare. Furthermore, children's ability to reason about addition concepts increased with age and problem‐solving proficiency. Reasoning about addition concepts was related to performance on the thematic, causal and visual reasoning tasks for older children but not for younger children. Overall, the findings suggest that while children's early knowledge of addition relations is domain specific, as children develop in their broader reasoning abilities these developments enhance their addition reasoning.

The Impact of Probability Judgments on Reasoning about Multiple Possibilities

HOROBIN, KAREN, and ACREDOLO, CURT. The Impact of Probability Judgments on Reasoning about Multiple Possibilities. CHILD DEVELOPMENT, 1989, 60, 183-200. When confronted with indeterminate problems, young children tend to select only 1 of the possible alternatives. The cause of this inappropriate closure was explored here in 2 studies in which children were administered random mixtures of single-solution and multiple-solution problems. A modified version of Pieraut-Le Bonniec's simple box problem was administered to 7-, 9-, and 12-year-old children, and a version of Acredolo and Horobin's complex relational reasoning task was administered to 9and 12-year-old children. In both studies, children were presented with the problems in both of 2 formats, 1 in which they were to list all possible solutions to each problem from a set of 3 alternatives, and a second in which they were to assign subjective probabilities to each of the 3 alternatives on a 5-point scale. Despite a strong tendency to close on single alternatives, most children correctly assigned nonzero probabilities to each of the possible alternatives. Furthermore, the tendency to acknowledge more than 1 alternative as possible correlated reliably with the tendency to assign equal as opposed to unequal probabilities to the possible alternatives. These results support the hypothesis that 1 major cause of inappropriate closure among young children is the confusion of possibility with probability.