Reasoning Tasks Research Articles

Dual-graph hierarchical interaction network for referring image segmentation

Referring Image Segmentation (RIS) aims to extract the object or stuff from an image according to the given natural language expression. As a representative multi-modal reasoning task, the main challenge of RIS lies in accurately understanding and aligning two types of heterogeneous data (i.e. image and text). Existing methods typically conduct this task via inexplicit cross-modal fusion toward the visual and linguistic features, which are separately extracted from different encoders and hard to capture accurate image–text alignment information due to their distinct latent representation structures. In this paper, we propose a Dual-Graph Hierarchical Interaction Network (DGHIN) to facilitate the explicit and comprehensive alignment between the image and text data. Firstly, two graphs are separately built for the initial visual and linguistic features extracted with different pre-trained encoders. By means of graph reasoning, we obtain a unified representation structure for different modalities to capture the intra-modal entities and their contexts, where each projected node incorporates the long-range dependencies into the latent representation. Then, the Hierarchical Interaction Module (HIM) is applied to the visual and linguistic graphs to extract comprehensive inter-modal correlations from the entity level and graph level, which not only capture the corresponding keywords and visual patches but also draws the whole sentence closer to the image region with the consistent context in the latent space. Extensive experiments on RefCOCO, RefCOCO+, G-Ref, and ReferIt demonstrate that the proposed DGHIN outperforms many state-of-the-art methods. Code is available at https://github.com/ZhaofengSHI/referring-DGHIN.

Do they have inquiry skill profiles? Exploring high school students' scientific inquiry in an animation‐based activity

AbstractIn this study, we aimed to characterize students' inquiry skill profiles and investigate whether students' gender, major, school location and type, and household registration are related to their inquiry skill profiles. By providing an animation‐based activity, we engaged students in a scientific inquiry on the atmospheric chemistry of climate change. Students performed data analytics, control of variables, and scientific reasoning tasks, which represented essential skills in the inquiry process. After removing the invalid data and conducting the two‐stage stratified sampling, we analyzed 724 11th‐grade Chinese students' multiple‐choice and open‐ended responses. A latent profile analysis (LPA) was used to identify if there were subgroups of students' inquiry skills. χ2 tests were conducted to examine whether the profiles' distribution differed in gender, major, school location, school type, and household registration. We identified four types of inquiry skill profiles among students: sophisticated, experimental, moderate, and basic, based on their skills in data analytics, scientific reasoning, and control of variables. The findings showed that school location significantly affected students' inquiry profiles, while school type, student major, and hukou had a mildly favorable impact. To sum up, the marriage of the LPA approach and the animation‐based activity has illuminated not just students' different inquiry skill profiles but also the relationships these profiles have with certain demographic factors. We discussed that it is imperative to recognize these varied inquiry skill profiles and work to bridge the disparities stemming from demographic differences for a more equitable science education environment.

Contrastive Video Question Answering via Video Graph Transformer.

We propose to perform video question answering (VideoQA) in a Contrastive manner via a Video Graph Transformer model (CoVGT). CoVGT's uniqueness and superiority are three-fold: 1) It proposes a dynamic graph transformer module which encodes video by explicitly capturing the visual objects, their relations and dynamics, for complex spatio-temporal reasoning. 2) It designs separate video and text transformers for contrastive learning between the video and text to perform QA, instead of multi-modal transformer for answer classification. Fine-grained video-text communication is done by additional cross-modal interaction modules. 3) It is optimized by the joint fully- and self-supervised contrastive objectives between the correct and incorrect answers, as well as the relevant and irrelevant questions respectively. With superior video encoding and QA solution, we show that CoVGT can achieve much better performances than previous arts on video reasoning tasks. Its performances even surpass those models that are pretrained with millions of external data. We further show that CoVGT can also benefit from cross-modal pretraining, yet with orders of magnitude smaller data. The results demonstrate the effectiveness and superiority of CoVGT, and additionally reveal its potential for more data-efficient pretraining.

Is it bad because it is boring? Effects of idle time on employee outcomes

ABSTRACT Idle time is a period during which employees cannot complete their work tasks for reasons beyond their control. It has previously been shown to have negative consequences for employees’ performance and well-being. To shed light on the mechanisms underlying these effects, we argue based on affective events theory and action regulation theory that idle time is an event that leads to boredom. Since boredom is a negative affective experience that motivates employees to do something about their situation, it may reduce job satisfaction and increase extra-role behaviour (i.e., prosocial and counterproductive) and turnover intentions. We tested our hypotheses using a sample of 1,036 employees who participated in a five-wave longitudinal study across 12 months. Results showed that, at both the within-person and between-person levels, idle time was positively associated with boredom, which, in turn, was associated with lower job satisfaction, higher counterproductive work behaviour, and higher turnover intentions, but not with prosocial work behaviour. These associations were robust when additionally controlling for work environment features (e.g., working from home), dispositions (e.g., personality), and demographic characteristics. Thus, idle time seems to be a boring and mostly negative experience for employees. Employees and managers should organize work to avoid boredom whenever possible.

Non-Axiomatic Logic Modeling of English Texts for Knowledge Discovery and Commonsense Reasoning

Non-axiomatic logic (NAL) is a term-based, non-monotonic, multi-valued logic with evidence-based formal semantics. All those characteristics position NAL as an excellent candidate for modeling natural language expressions and supporting artificial agents while performing knowledge discovery and commonsense reasoning tasks. In this article, we propose a set of rules for the automatic translation of natural language (NL) text into the formal language of non-axiomatic logic (NAL). Several free available tools are used to support a previous linguistic analysis, and a common sense ontology is used to populate a background knowledge base that helps to delimit the scope and the semantics of logical formulas translated. Experimentation shows our set to be the most comprehensive NL-to-NAL translation rule set known so far. Furthermore, we included an extensive set of examples to show how our proposed set of rules can be used for translating a wide range of English statements with varying grammatical structures.

Language models and psychological sciences.

Large language models (LLMs) are demonstrating impressive performance on many reasoning and problem-solving tasks from cognitive psychology. When tested, their accuracy is often on par with average neurotypical adults, challenging long-standing critiques of associative models. Here we analyse recent findings at the intersection of LLMs and cognitive science. Here we discuss how modern LLMs resurrect associationist principles, with abilities like long-distance associations enabling complex reasoning. While limitations remain in areas like causal cognition and planning, phenomena like emergence suggest room for growth. Providing examples and increasing the dimensions of the network are methods that further improve LLM abilities, mirroring facilitation effects in human cognition. Analysis of LLMs errors provides insight into human cognitive biases. Overall, we argue LLMs represent a promising development for cognitive modelling, enabling new explorations of the mechanisms underlying intelligence and reasoning from an associationist point of view. Carefully evaluating LLMs with the tools of cognitive psychology will further understand the building blocks of the human mind.

Cooperative task allocation for multi heterogeneous aerial vehicles using particle swarm optimization algorithm and entropy weight method

Reasonable and effective task allocation strategy is the premise to ensure the successful completion of combat missions, which is of great significance to improve the overall operational effectiveness of the cluster. In this paper, a cooperative task allocation method for multi heterogeneous aerial vehicles under strong space-time constraints using particle swarm optimization (PSO) algorithm and entropy weight method is presented. Specifically, the problem of cooperative task allocation is formulated as a multi-objective optimization problem under multi-constraint conditions, and the requirement of required time of arrival (RTA) of time-sensitive targets is considered. Then, the PSO algorithm is applied to solve the task allocation model. Especially, task sequence is introduced into the coding of the feasible solutions, and task urgency is employed to update the particle position. Furthermore, the entropy weight method is introduced into task assignment scheme decision-making. Finally, a simulation is given to illustrate the validity.

Winning or losing: Children’s proportional reasoning across motivational contexts

Children tend to prioritize whole number information over relational information in proportional reasoning tasks, such that they judge a spinner with 4/10 red pieces as more likely to land on red than a spinner with 2/3 red pieces, because 4 > 2 (e.g., Hurst & Cordes, 2018a; Jeong et al., 2007). This whole number bias is hypothesized to be a driven by fluency in verbal counting in early childhood, which is thought to promote attention to whole number information. In this study, we explored (1) the relation between verbal counting abilities and whole number biases and (2) whether distinct framing contexts – either encouraging children to maximize the number of stickers won, or minimizing the number of stickers lost - differentially impact children’s proportional reasoning. Three- to nine-year olds (N = 210, Mage = 5.7 years) chose which of two spinners they preferred to spin. Children in the Gain condition learned that if the spinner landed on red, they would win a sticker and if it landed on blue, nothing would happen; children in the Loss condition learned that if the spinner landed on red, nothing would happen, but if it landed on blue, they would lose a sticker. Counter to prior work, performance of both older (6–9 year olds) and younger (3–5 year olds) children revealed whole number biases. Notably, whole number biases were not related to counting abilities. Importantly, we find framing the task in a Loss scenario lessened whole number biases, suggesting that task framing can alter children’s attention to whole number information in a proportional reasoning context.

How Time Pressure in Different Phases of Decision-Making Influences Human-AI Collaboration

Human cognitive and decision-making abilities depreciate under pressure, motivating the emergence of artificial intelligence (AI) systems as decision support tools to assist people in performing tasks under stress. In this work, we study human decision-making behavior and task performance under time pressure---induced from limitedinitial observation time (time to perform the task before providing an initial response without AI input) andfinal decision time (time to weigh an AI's suggestion before reaching a collective human-AI team answer)---for spatial reasoning and count estimation tasks. Our results show that, while the impact of initial observation time on AI-assisted decision-making was dependent on task nature, participants were more likely to follow AI suggestions when they were provided with longer final decision time; moreover, although participants generally tended to adhere to their initial responses, they had more agency when they were more logically engaged in a task. Our results offer a nuanced understanding of human-AI collaboration under time pressure in different phases of the decision-making process.

A Differential Datalog Interpreter

The core reasoning task for datalog engines is materialization, the evaluation of a datalog program over a database alongside its physical incorporation into the database itself. The de-facto method of computing is through the recursive application of inference rules. Due to it being a costly operation, it is a must for datalog engines to provide incremental materialization; that is, to adjust the computation to new data instead of restarting from scratch. One of the major caveats is that deleting data is notoriously more involved than adding since one has to take into account all possible data that has been entailed from what is being deleted. Differential dataflow is a computational model that provides efficient incremental maintenance, notoriously with equal performance between additions and deletions, and work distribution of iterative dataflows. In this paper, we investigate the performance of materialization with three reference datalog implementations, out of which one is built on top of a lightweight relational engine, and the two others are differential-dataflow and non-differential versions of the same rewrite algorithm with the same optimizations. Experimental results suggest that monotonic aggregation is more powerful than ascenting merely the powerset lattice.

A task matching model of photovoltaic storage system under the energy blockchain environment - based on GA-CLOUD-GS algorithm

Photovoltaic storage system (PVSS) has been spawned with the combined application of photovoltaic (PV), energy storage (ES) and energy blockchain (EB), which has also made important contributions to the energy structure adjustment, energy transaction security and ecological environment protection. The establishment of a reasonable task matching mechanism can further improve the operation efficiency and quality of the system under the EB environment, and this paper consequently constructs a task matching model of PVSS based on GA-CLOUD-GS algorithm. Firstly, the decision-making committee is set up and the index system is determined for the to-be-matched task. Secondly, the Genetic Algorithm (GA) comprehensive weighting method is introduced to integrate the subjective and objective weights to achieve the weight optimization. Thirdly, the cloud model is used to obtain the preference order and priority order of the to-be-matched subjects. Fourthly, the Gale-Shapley (GS) algorithm is used to obtain the task matching results. Finally, the operability, the stability and the effectiveness of the model are separately verified through numerical simulation, sensitivity analysis and comparative analysis. Corresponding suggestions for the task matching of PVSS in the EB environment are also put forward according to the research content.

Elevated CRP and TNF-α levels are associated with blunted neural oscillations serving fluid intelligence

IntroductionInflammatory processes help protect the body from potential threats such as bacterial or viral invasions. However, when such inflammatory processes become chronically engaged, synaptic impairments and neuronal cell death may occur. In particular, persistently high levels of C-reactive protein (CRP) and tumor necrosis factor-alpha (TNF-α) have been linked to deficits in cognition and several psychiatric disorders. Higher-order cognitive processes such as fluid intelligence (Gf) are thought to be particularly vulnerable to persistent inflammation. Herein, we investigated the relationship between elevated CRP and TNF-α and the neural oscillatory dynamics serving Gf. MethodsSeventy adults between the ages of 20–66 years (Mean = 45.17 years, SD = 16.29, 21.4% female) completed an abstract reasoning task that probes Gf during magnetoencephalography (MEG) and provided a blood sample for inflammatory marker analysis. MEG data were imaged in the time–frequency domain, and whole-brain regressions were conducted using each individual’s plasma CRP and TNF-α concentrations per oscillatory response, controlling for age, BMI, and education. ResultsCRP and TNF-α levels were significantly associated with region-specific neural oscillatory responses. In particular, elevated CRP concentrations were associated with altered gamma activity in the right inferior frontal gyrus and right cerebellum. In contrast, elevated TNF-α levels scaled with alpha/beta oscillations in the left anterior cingulate and left middle temporal, and gamma activity in the left intraparietal sulcus. DiscussionElevated inflammatory markers such as CRP and TNF-α were associated with aberrant neural oscillations in regions important for Gf. Linking inflammatory markers with regional neural oscillations may hold promise in identifying mechanisms of cognitive and psychiatric disorders.

GPT-3.5, GPT-4, or BARD? Evaluating LLMs reasoning ability in zero-shot setting and performance boosting through prompts

Large Language Models (LLMs) have exhibited remarkable performance on various Natural Language Processing (NLP) tasks. However, there is a current hot debate regarding their reasoning capacity. In this paper, we examine the performance of GPT-3.5, GPT-4, and BARD models, by performing a thorough technical evaluation on different reasoning tasks across eleven distinct datasets. Our paper provides empirical evidence showcasing the superior performance of ChatGPT-4 in comparison to both ChatGPT-3.5 and BARD in zero-shot setting throughout almost all evaluated tasks. While the superiority of GPT-4 compared to GPT-3.5 might be explained by its larger size and NLP efficiency, this was not evident for BARD. We also demonstrate that the three models show limited proficiency in Inductive, Mathematical, and Multi-hop Reasoning Tasks. To bolster our findings, we present a detailed and comprehensive analysis of the results from these three models. Furthermore, we propose a set of engineered prompts that enhances the zero-shot setting performance of all three models.

Visual Reasoning: From State to Transformation.

Most existing visual reasoning tasks, such as CLEVR in VQA, ignore an important factor, i.e., transformation. They are solely defined to test how well machines understand concepts and relations within static settings, like one image. Such state driven visual reasoning has limitations in reflecting the ability to infer the dynamics between different states, which has shown to be equally important for human cognition in Piaget's theory. To tackle this problem, we propose a novel transformation driven visual reasoning (TVR) task. Given both the initial and final states, the target becomes to infer the corresponding intermediate transformation. Following this definition, a new synthetic dataset namely TRANCE is first constructed on the basis of CLEVR, including three levels of settings, i.e., Basic (single-step transformation), Event (multi-step transformation), and View (multi-step transformation with variant views). Next, we build another real dataset called TRANCO based on COIN, to cover the loss of transformation diversity on TRANCE. Inspired by human reasoning, we propose a three-staged reasoning framework called TranNet, including observing, analyzing, and concluding, to test how recent advanced techniques perform on TVR. Experimental results show that the state-of-the-art visual reasoning models perform well on Basic, but are still far from human-level intelligence on Event, View, and TRANCO. We believe the proposed new paradigm will boost the development of machine visual reasoning. More advanced methods and new problems need to be investigated in this direction.

Handicraft art leisure activities and cognitive reserve

ObjectiveOlder individuals face a higher likelihood of developing dementia. The rate of cognitive decline resulting from dementia is not equivalent for all, as some patients with dementia are able to function independently longer than others, despite having similar disease burden. The cognitive reserve (CR) theory provides one explanation for the differing rate of decline. CR suggests that there are factors—most notably, educational attainment and occupational attainment—that can protect against the cognitive decline. Although the beneficial effects of these notable CR factors are clear, not all are easily modifiable. Participation in leisure activities may represent a more easily modifiable factor. Some research hints at beneficial effects of leisure activities, although specific leisure activities have not been well examined. The present study examined the relations between handicraft art leisure activities (HALAs) and multiple cognitive domains. MethodArchival WAIS-IV and demographic data for 50 California retirement community residents were examined. ResultsHALA participation accounted for statistically significant variance in working memory performance (R 2 = .40, β = .24%) over and above the established CR factors of age, depression, educational attainment, and occupational attainment. In addition, HALA participation was related to a better ability to perform abstract visual information tasks (Block Design subtest, r = .28, p = .05) and non-verbal reasoning tasks (Visual Puzzles subtest, r = .38, p = .008). ConclusionsHALA participation among older adults could contribute to the retention of cognitive function, supporting the role of HALA participation as a CR factor.

We Cannot Ignore the Signs: The Development of Equivalence and Arithmetic for Students from Grades 3 to 4

ABSTRACT Students' understanding of the meaning of the equal sign develops slowly over the primary grades. In addition to updating their representations of equations to recognize that the equal sign represents an equivalence relation rather than signaling an operation, students need to move beyond full computation to efficiently solve equivalence problems. In this study, we examined the longitudinal relation between arithmetic and equivalence for students who were capable of accurately solving arithmetic problems in different formats. Chinese students (N = 612; M age = 9.0 years in Grade 3, 57% boys) completed measures of arithmetic fluency and equivalence fluency in Grade 3 and again in Grade 4. They also completed a non-verbal reasoning task in Grade 3. We tested a cross-lagged structural equation model to examine the reciprocal relations between arithmetic and equivalence fluency. We found reciprocal relations between the development of arithmetic and equivalence fluency from Grades 3 to 4, with a greater influence of arithmetic on the development of equivalence than the reverse. Furthermore, non-verbal reasoning predicted the development of equivalence, but not the development of arithmetic. Based on our findings, we conclude that for Chinese students with prior basic understanding of equivalence, flexible access to arithmetic facts supports their development of equivalence fluency.

Zero-shot visual reasoning through probabilistic analogical mapping

Human reasoning is grounded in an ability to identify highly abstract commonalities governing superficially dissimilar visual inputs. Recent efforts to develop algorithms with this capacity have largely focused on approaches that require extensive direct training on visual reasoning tasks, and yield limited generalization to problems with novel content. In contrast, a long tradition of research in cognitive science has focused on elucidating the computational principles underlying human analogical reasoning; however, this work has generally relied on manually constructed representations. Here we present visiPAM (visual Probabilistic Analogical Mapping), a model of visual reasoning that synthesizes these two approaches. VisiPAM employs learned representations derived directly from naturalistic visual inputs, coupled with a similarity-based mapping operation derived from cognitive theories of human reasoning. We show that without any direct training, visiPAM outperforms a state-of-the-art deep learning model on an analogical mapping task. In addition, visiPAM closely matches the pattern of human performance on a novel task involving mapping of 3D objects across disparate categories.

Metarules, Judgment and the Algorithmic Future of Financial Regulation in the UK

Abstract UK financial regulators are experimenting with the conversion of rulebook content into machine-readable and executable code. A major driver of these initiatives is the belief that the use of algorithms will eliminate the need for human interpretation as a deliberative process, and that this would be a welcome development because it will improve effectiveness while cutting time and costs for regulators and the industry alike. In this article, I set out to explain why human interpretation should be preserved and further harnessed if data-driven governance is to work at all. To support my thesis, I bring attention to the limited translatability of rulebook content into code, and to the difficulties for machines to engage with the full spectrum of tasks of analogical reasoning. I further contend that it would be desirable to preserve human interpretation on procedural grounds pertaining to the legitimacy of financial regulators. I conclude with recommendations about the future design of the financial rulebooks.

Neural activity and network analysis for understanding reasoning using the matrix reasoning task.

Reasoning requires the ability to manipulate mental representations and understand relationships between objects. There is a paucity of research regarding the functional connections between multiple brain areas that may interact during commonly used reasoning tasks. The present study aimed to examine functional activation and connectivity of frontoparietal regions during a Matrix Decision Making Task, completed by twenty-one right-handed healthy participants while undergoing fMRI. Voxel-wise whole brain analysis of neural response to the task revealed activation spanning dorsal and lateral prefrontal, occipital, and parietal regions. Utilizing Group Iterative Multiple Model Estimation, a data-driven approach that estimates the presence and direction of connectivity between specific ROIs, connectivity between prefrontal and sensory processing regions were revealed. Moreover, the magnitude of connectivity strength between the left precentral gyrus and left dorsal cingulate (dACC) was positively correlated with MR behavioral performance. Taken together, results are consistent with earlier work demonstrating involvement of regions comprising the central executive network in relational reasoning. These data expand existing knowledge regarding communication of key brain regions during the task and demonstrate that understanding how key brain regions are interconnected can effectively predict the quality of behavioral output.

Is adolescence believed to be a period of greater risk taking than adulthood?

ABSTRACT Little is known about adolescents’ own beliefs regarding their level of risk taking or regarding peer influence on the latter. This is an important matter given that beliefs influence judgements and decisions. With the present study, we aimed to study adolescents’ and adults’ beliefs about adolescents’ risk taking compared to adults’ risk taking, and beliefs about peer influence on risk taking. To this end, an experimental design was used. A cross-sectional study included 56 adolescents and 43 adults, who completed a questionnaire asking about adolescents’ and adults’ risk-taking propensity and a syllogistic reasoning task designed to indirectly study these beliefs. Both direct and indirect measures indicated that adolescents perceived adolescence as a period of higher risk taking compared to adulthood, and believed peer presence to promote this effect. Adults perceived this detrimental effect of peers to be present irrespective of age. We discuss the implications of these results in terms of social representation.