Search

In recent years, the establishment of a substantial amount of academic groups on scientific social network has brought new opportunities for the collaboration among researchers. In this situation, conducting paper recommendation to these academic groups is of terrific necessity in that it can further facilitate group activities. However, when producing group recommendation, existing methods fail to make full use of the abundant group information, from which a great deal of valuable information can be inferred to facilitate the recommendation performance. In addition, those methods tend to assign an equal weight to each group member when aggregating their recommendations, which is unreasonable in practice. Although some improvements have been made to remedy this problem by assigning different weights to group members, they fail to take into account the reliabilities of group members. Therefore, a group-oriented paper recommendation method based on probabilistic matrix factorization and evidential reasoning (GPMF_ER) is proposed in this article to tackle these problems. More specifically, the group and paper content information are integrated into the probabilistic matrix factorization model to enhance the accuracy of individual recommendation. Afterward, evidential reasoning rule is introduced in the aggregation step to consider both the weights and reliabilities of group members. Extensive experiments have been conducted on the real world CiteULike dataset and the results demonstrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Promoting a wider range of contribution types can facilitate healthy growth of the visualization community, while increasing the intellectual diversity of visualization research papers. In this paper, we discuss the importance of contribution types and summarize contribution types that can be meaningful in visualization research. We also propose several concrete next steps we can and should take to ensure a successful launch of the contribution types. [ABSTRACT FROM AUTHOR]

This work analyzes sensemaking frameworks and experiments with an iteratively designed visual analysis tool to identify design implications for facilitating research idea generation using visualizations. Our tool, ThoughtFlow, structures and visualizes literature collections using topic models to bridge the information gap between core activities during research ideation. To help users stay focused on a topic while discovering relevant documents, we designed and analyzed usage patterns for two types of embedded visualization that help determine document relevance while minimizing distraction. We analyzed how research ideation outcomes and processes differ when using ThoughtFlow and conventional search engines by augmenting insight-based evaluation with concept-map analysis. Our results suggest that operations afforded by topic models match well with later ideation stages when coherent topics have emerged, but not with early stages when users are still relying heavily on individual keywords to gather background knowledge. We also present qualitative evidence that citation sparklines encourage more exploration of recommended references, and that a preference for paper thumbnails may depend on the consistency between the evidence and the current mental frame. [ABSTRACT FROM AUTHOR]

Electroencephalogram (EEG) signals are widely utilized in the field of cognitive workload decoding (CWD). However, when the recognition scenario is shifted from subject-dependent to subject-independent or spans a long period, the accuracy of CWD deteriorates significantly. Current solutions are either dependent on extensive training datasets or fail to maintain clear distinctions between categories, additionally lacking a robust feature extraction mechanism. In this paper, we tackle these issues by proposing a Bi-Classifier Joint Domain Adaptation (BCJDA) model for EEG-based cross-time and cross-subject CWD. Specifically, the model consists of a feature extractor, a domain discriminator, and a Bi-Classifier, containing two sets of adversarial processes for domain-wise alignment and class-wise alignment. In the adversarial domain adaptation, the feature extractor is forced to learn the common domain features deliberately. The Bi-Classifier also fosters the feature extractor to retain the category discrepancies of the unlabeled domain, so that its classification boundary is consistent with the labeled domain. Furthermore, different adversarial distance functions of the Bi-Classifier are adopted and evaluated in this model. We conduct classification experiments on a publicly available BCI competition dataset for recognizing low, medium, and high cognitive workload levels. The experimental results demonstrate that our proposed BCJDA model based on cross-gradient difference maximization achieves the best performance.

Smart cyber-physical systems (S-CPSs) are complex engineered systems empowered by cyber-physical computing and equipped with the capability of reasoning, learning, adapting, and evolving. As an outcome of data-driven dynamic computing, reasoning capabilities, and the run-time obtained own knowledge, nonlinear and emergent behavior of S-CPSs whilst in operation is an open issue, not experienced in the case of conventional technical systems. This paper analyzes the technical issues of run-time operation and emergent behavior of S-CPSs, reviews the current understanding and state of advancement in designing S-CPSs for run-time, explores the paradox, and issues of designing for run-time adaptation, and synthesizes some general principles that can be taken into consideration when addressing the challenges, first of all, in the context of advanced manufacturing systems. This paper introduces four levels of CPSs according to reasoning capabilities and adaptation freedom of systems, and recognizes the paradox that a system with a higher level of freedom requires a higher level of self-control and resource management according to the overall objective of operation. Specific and common design principles are presented and critically assessed for each advancement level of CPSs. The principles synthesized by the authors provide only a partial fulfillment of the generic need. The planned future research addresses these issues and proposes (largely implementation and application independent) genuine principles for system developers. [ABSTRACT FROM AUTHOR]

The eight papers in this special issue focus on cognitive inspired networked applications. Cognition is emerging as a new and promising methodology with the development of cognitive-inspired computing, ubiquitous networks, and systems, which has the potential to enable a large class of applications. However, recent advances, such as interference alignment, collaborative communications, and cognitive computational theory, make us lack a clear understanding of analytical cognitive-inspired computing, networks, and systems, as well as best practices, to design resource-efficient, utility, pervasive, scalable, and autonomy oriented cognitive systems and applications. The objective of this special Issue is to bring together state-of-the-art research contributions that address these key aspects of cognitive inspired network systems and applications. [ABSTRACT FROM PUBLISHER]

Quality inspection, a widely adopted practice in supply chains, measures whether delivered products conform to prespecified quality requirements. Due to potential economic benefits, suppliers may deliberately manipulate products to falsify test results. However, unqualified products could cause severe problems in supply chains or even tragedies, such as China’s tainted milk scandal. In this paper, we propose a belief-desire-intention modeling approach to predict suppliers’ behavior and provide inspection suggestions to buyers to overcome the problem. Assuming access to the production process, our approach can represent suppliers’ knowledge of production and deception to mimic their reasoning processes and predict their deception intentions. This flexible approach can also adapt to environmental changes and deliver effective results. In this paper, we build a prototype system for supply chain quality inspections based on the proposed method. We conduct laboratory experiments to collect data for computational assessments of the performance of the prototype. It is shown that our proposed approach is more accurate than classic machine learning methods in detecting suppliers’ deceptions. [ABSTRACT FROM AUTHOR]

A smart home needs to be human-centric, where it tries to fulfill human needs given the devices it has. Various works are developed to provide homes with reasoning and planning capability to fulfill goals, but most do not support complex sequence of plans or require significant manual effort in devising subplans. This is further aggravated by the need to optimize conflicting personal goals. A solution is to solve the planning problem represented as constraint satisfaction problem (CSP). But CSP uses hard constraints and, thus, cannot handle optimization and partial goal fulfillment efficiently. This paper aims to extend this approach to weighted CSP. Knowledge representation to help in generating planning rules is also proposed, as well as methods to improve performances. Case studies show that the system can provide intelligent and complex plans from activities generated from semantic annotations of the devices, as well as optimization to maximize personal constraints’ fulfillment. Note to Practitioners—Smart home should maximize the fulfillment of personal goals that are often conflicting. For example, it should try to fulfill as much as possible the requests made by both the mother and daughter who wants to watch TV but both having different channel preferences. That said, every person has a set of goals or constraints that they hope the smart home can fulfill. Therefore, human-centric system that automates the loosely coupled devices of the smart home to optimize the goals or constraints of individuals in the home is developed. Automated planning is done using converted services extracted from devices, where conversion is done using existing tools and concepts from Web technologies. Weighted constraint satisfaction that provides the declarative approach to cover large problem domain to realize the automated planner with optimization capability is proposed. Details to speed up planning through search space reduction are also given. Real-time case studies are run in a prototype smart home to demonstrate its applicability and intelligence, where every planning is performed under a maximum of 10 s. The vision of this paper is to be able to implement such system in a community, where devices everywhere can cooperate to ensure the well-being of the community. [ABSTRACT FROM AUTHOR]

Contribution: This paper illustrates how to design and implement curricula in terms of the combined use of flipped learning and inquiry-based learning in an engineering course. Background: Elementary courses in engineering schools are conventional and foundational, and involve a considerable amount of knowledge. Throughout such courses, students are also expected to develop insight, which cannot be obtained by only listening to instructors. Having relevant discussions is also difficult for most instructors. Intended Outcomes: The combined use of flipped learning and inquiry-based learning would be beneficial to broaden student achievement. Application Design: Based on an epistemological approach about knowledge and knowing, this paper applies the combined use of flipped learning and inquiry-based learning to enhance student knowledge and advance ways of thinking on a System Modeling and Control course. Findings: The extended learning time and the collective responsibility for learning are discussed as critical issues in applying the combined use of flipped learning and inquiry-based learning in an engineering school. [ABSTRACT FROM AUTHOR]

Even though digital learning games have become common in education, relatively little is known about the usefulness of game-based assessment. This paper aims to explore if a game-based math test can provide added value to math education with respect to cognitive and affective outcomes. We used in-game measures, embedded in the game called Semideus Exam, focusing on conceptual fraction knowledge. In order to validate the game-based assessment approach, we compared the cognitive outcomes of 51 Finnish sixth graders, who completed both paper-based and game-based math tests in a randomized order. In addition, the students’ test anxiety and flow experience were measured to evaluate the affective outcomes. The results indicate that the game-based test scores correlated significantly with the paper-based test scores suggesting that the game-based assessment was successfully implemented and the game provided comparable data with the paper-based test approach. More importantly, the results revealed that game-based assessment lowered test anxiety and increased engagement which is likely to decrease assessment bias caused by test anxiety. In addition, the results show that earlier playing experience and gender did not influence the game-based test score suggesting fairness of the game-based assessment approach. Although we identified several benefits of the game-based assessment approach, more evidence is needed on the usefulness and fairness of game-based assessments. [ABSTRACT FROM AUTHOR]

Emotion Cause Extraction (ECE) aims to reveal the cause clauses behind a given emotion expressed in a text, which has become an emerging topic in broad research communities, such as affective computing and natural language processing. Despite the fact that current methods about the ECE task have made great progress in text semantic understanding from lexicon- and sentence-level, they always ignore the certain causal narratives of emotion text. Significantly, these causal narratives are presented in the form of semantic structure and highly helpful for structure-level emotion cause understanding. Nevertheless, causal narrative is just an abstract narratological concept and its involving semantics is quite different from the common sequential information. Thus, how to properly model and utilize such particular narrative information to boost the ECE performance still remains an unresolved challenge. To this end, in this paper, we propose a novel Causal Narrative Comprehension Model (CNCM) for emotion cause extraction, which learns and leverages causal narrative information smartly to address the above problem. Specifically, we develop a Narrative-aware Causal Association (NCA) unit, which mines the narrative cue about emotional results and uses the semantic correlation between causes and results to model causal narratives of documents. Besides, we design a Result-aware Emotion Attention (REA) unit to make full use of the known result of causal narrative for multiple understanding about emotional causal associations. Through the ingenious combination and collaborative utilization of these two units, we could better identify the emotion cause in the text with causal narrative comprehension. Extensive experiments on the public English and Chinese benchmark datasets of ECE task have validated the effectiveness of CNCM with significant margin by comparing with the state-of-the-art baselines, which demonstrates the potential of narrative information in long text understanding. [ABSTRACT FROM AUTHOR]

Given a natural language expression and an image/video, the goal of referring segmentation is to produce the pixel-level masks of the entities described by the subject of the expression. Previous approaches tackle this problem by implicit feature interaction and fusion between visual and linguistic modalities in a one-stage manner. However, human tends to solve the referring problem in a progressive manner based on informative words in the expression, i.e., first roughly locating candidate entities and then distinguishing the target one. In this paper, we propose a cross-modal progressive comprehension (CMPC) scheme to effectively mimic human behaviors and implement it as a CMPC-I (Image) module and a CMPC-V (Video) module to improve referring image and video segmentation models. For image data, our CMPC-I module first employs entity and attribute words to perceive all the related entities that might be considered by the expression. Then, the relational words are adopted to highlight the target entity as well as suppress other irrelevant ones by spatial graph reasoning. For video data, our CMPC-V module further exploits action words based on CMPC-I to highlight the correct entity matched with the action cues by temporal graph reasoning. In addition to the CMPC, we also introduce a simple yet effective Text-Guided Feature Exchange (TGFE) module to integrate the reasoned multimodal features corresponding to different levels in the visual backbone under the guidance of textual information. In this way, multi-level features can communicate with each other and be mutually refined based on the textual context. Combining CMPC-I or CMPC-V with TGFE can form our image or video version referring segmentation frameworks and our frameworks achieve new state-of-the-art performances on four referring image segmentation benchmarks and three referring video segmentation benchmarks respectively. Our code is available at https://github.com/spyflying/CMPC-Refseg. [ABSTRACT FROM AUTHOR]

This paper presents an appliance identification algorithm for use with a non-intrusive home energy monitor based on a cogent confabulation neural network. As a cogent confabulation neural network does not require multiplications during the identification phase, it is an effective choice for systems with low-computational capability. A non-intrusive home energy monitor needs to learn not only the energy patterns of individual appliances but also those of combinations of appliances. To relieve the burden of learning power patterns of the combinations, this paper proposes a parameter-building scheme based on the parameters of individual appliances. The proposed algorithm is evaluated on datasets prepared by the reference energy disaggregation dataset and the authors. The average success rate was 83.8% for up to eight appliances and showed better performance than the combinatorial optimization and artificial neural network approaches. [ABSTRACT FROM AUTHOR]

Corporate meetings are a crucial part of business activities. While numerous academic papers investigated how to make the scheduling process of meetings faster or even automatic, little work has been done yet to facilitate the retrospective reasoning about how time is spent on meetings. Traditional calendar applications do not allow users to extract actionable statistics although it has been shown that reflection-oriented design can increase the users’ understanding of their habits and can thereby encourage a shift towards better practices. In this paper, we present MineTime Insight, a tool made of multiple coordinated views for the exploration of personal calendar data, with the overarching goal of improving short and long-term scheduling decisions. Despite being focused on the working environment, our work builds upon recent results in the field of Personal Visual Analytics, as it targets users not necessarily expert in visualization and data analysis. We demonstrate the potential of MineTime Insight, when applied to the agenda of an executive manager. Finally, we discuss the results of an informal user study and a field study. Our results suggest that our visual representations are perceived as easy to understand and helpful towards a change in the scheduling habits. [ABSTRACT FROM AUTHOR]

This paper addresses the task of detecting and recognizing human-object interactions (HOI) in images. Considering the intrinsic complexity and structural nature of the task, we introduce a cascaded parsing network (CP-HOI) for a multi-stage, structured HOI understanding. At each cascade stage, an instance detection module progressively refines HOI proposals and feeds them into a structured interaction reasoning module. Each of the two modules is also connected to its predecessor in the previous stage, enabling efficient cross-stage information propagation. The structured interaction reasoning module is built upon a graph parsing neural network (GPNN), which efficiently models potential HOI structures as graphs and mines rich context for comprehensive relation understanding. In particular, GPNN infers a parse graph that i) interprets meaningful HOI structures by a learnable adjacency matrix, and ii) predicts action (edge) labels. Within an end-to-end, message-passing framework, GPNN blends learning and inference, iteratively parsing HOI structures and reasoning HOI representations (i.e., instance and relation features). Further beyond relation detection at a bounding-box level, we make our framework flexible to perform fine-grained pixel-wise relation segmentation; this provides a new glimpse into better relation modeling. A preliminary version of our CP-HOI model reached 1st place in the ICCV2019 Person in Context Challenge, on both relation detection and segmentation. In addition, our CP-HOI shows promising results on two popular HOI recognition benchmarks, i.e., V-COCO and HICO-DET. [ABSTRACT FROM AUTHOR]

In recent years, weakly supervised object detection has attracted great attention in the computer vision community. Although numerous deep learning-based approaches have been proposed in the past few years, such an ill-posed problem is still challenging and the learning performance is still behind the expectation. In fact, most of the existing approaches only consider the visual appearance of each proposal region but ignore to make use of the helpful context information. To this end, this paper introduces two levels of context into the weakly supervised learning framework. The first one is the proposal-level context, i.e., the relationship of the spatially adjacent proposals. The second one is the semantic-level context, i.e., the relationship of the co-occurring object categories. Therefore, the proposed weakly supervised learning framework contains not only the cognition process on the visual appearance but also the reasoning process on the proposal- and semantic-level relationships, which leads to the novel deep multiple instance reasoning framework. Specifically, built upon a conventional CNN-based network architecture, the proposed framework is equipped with two additional graph convolutional network-based reasoning models to implement object location reasoning and multi-label reasoning within an end-to-end network training procedure. Comprehensive experiments on the widely used PASCAL VOC and MS COCO benchmarks have been implemented, which demonstrate the superior capacity of the proposed approach when compared with other state-of-the-art methods and baseline models. [ABSTRACT FROM AUTHOR]

Negation is a new perspective to represent knowledge. The negation of probability distribution has been proposed, and it has a lot of interesting properties, which can reach a maximum entropy. Because of the defects of the classical probability theory in the expression of uncertainty, the basic belief assignment (BBA) in the Dempster–Shafer theory (D–S theory) are widely used in decision theory. Thus, negation provides a new perspective for D–S theory to measure fuzziness. In this paper, a new definition of negation of BBA is presented. In the proposed negation, BBAs are represented as vectors, and negation is realized by matrix operators. This method has a good interpretation of the matrix operators and has the merit of simplifying the problem. With several different definitions of entropy to determinate the uncertainty of BBA, the proposed negation of BBA can reach a maximum belief entropy when the entropies satisfy a certain property. [ABSTRACT FROM AUTHOR]

Bodies of knowledge (BOK) contain relevant knowledge for several areas. Knowledge representation for BOK is essential to understand the true context and the possible application of BOK in science. Knowledge can be symbolically represented in many ways, and in this paper graphs and software design patterns were chosen to develop knowledge representation techniques for BOK. [ABSTRACT FROM AUTHOR]

Intelligent systems are desired in dynamic fault diagnoses for large and complex systems such as nuclear power plants. Dynamic uncertain causality graph (DUCG) is such a system presented previously. This paper extends the DUCG methodology to deal with negative feedbacks, which is one of the most difficult problems in fault diagnosis, and predicts the fault development online. Two methods are presented. One does not involve causality propagation across time slices. Another involves causality propagation through time slices. A nuclear power plant simulator located at Tsinghua University is used to test the DUCG methodology. A typical experiment involving negative feedback is given, which jointly applies the methods separately presented in different papers, such as how to deal with dynamic fault diagnosis and directed cyclic graphs (DCGs), and the methods presented in this paper. Results show that DUCG is powerful in knowledge representation, diagnosing possible faults, and predicting developments of faults. It also demonstrates that DUCG is robust, which means that the DUCG inference does not rely much on the accuracy of probability parameters. The logics of diagnoses and predictions are graphically displayed, so that users know not only inference results, but also why they are correct. [ABSTRACT FROM AUTHOR]

This paper presents track fusion and behavioral reasoning for moving vehicles in close proximity based on the curvilinear coordinates of roadway geometries. The inferred track and behavior of other vehicles can be used to perform safe actions in intelligent vehicle applications and autonomous driving. Vehicle detections from multiple perception sensors are integrated using track-to-track (T2T) fusion based on a cross-covariance method, and this T2T fusion is performed with curvilinear coordinates created using prebuilt roadway geometry on a digital map. The coordinate conversion to curvilinear space has many benefits for behavioral reasoning and tracking, such as constraining problem spaces and dimensions. A machine learning classifier based on a support vector machine is then applied to deduce the behavior of nearby vehicles. The algorithms presented here for track fusion and behavioral reasoning based on curvilinear coordinates have been verified through experiments in various real traffic scenarios. [ABSTRACT FROM AUTHOR]

This paper presents the results from a one-year study of 12 patients with moderate dementia in an adult day program who played a novel whack-a-mole game-based measurement instrument for cognitive behavior and performance. The ongoing measurement of cognition and changes associated with dementia is a challenge for healthcare providers. Measurement methods based on a tablet-based instrument are proposed. Partnership with the adult day program greatly eased recruitment: all but 1 eligible participant joined our study, compared to one in five, or lower, for previous studies with similar populations. There are three unique aspects to the design of our game: first, it has two distinct targets requiring different actions, which increases the cognitive processing for the users; second, each level is systematically more difficult; third, it records and analyzes player performance. The results show that the patients’ game performance improves over the first few weeks; this indicates that they are learning the game and retaining ability gains from week-to-week, suggesting some procedural learning is still intact. Over the year, 4 participants showed cognitive decline, 4 were stable and 3 improved based on their Minimental State Exam (MMSE) score. Two measures are proposed based on level progression within the sessions and mole hit performance. The level progression measure identifies declining participants with one false negative (FN) and one false positive error. The mole hit performance measure identifies declining participants with one FN error. These results demonstrate the potential for the proposed instrument to provide an ongoing measurement as an alternative for the repeated application of the MMSE. [ABSTRACT FROM AUTHOR]

Usually, the evaluation of the classifiers performance is not an easy task to be performed, mainly when we analyze different criteria (output parameters). In this evaluation process, we can use quantitative measures (accuracy, specificity, among others), however, when the output values are very close and we have several criteria, the results are difficult to be interpreted by users. This paper aims to propose a new linguistic model to evaluate the performance of several classifiers. It is based on the notion of gradual complex numbers (GCN), proposed in [18]. In this paper, we present the theoretical basis of GCNs for classifier evaluator and we assess the performance of the proposed model (GCN) through an empirical study. In addition, the performance of GCN is compared with that of fuzzy complex numbers [6], and it reveals gains. [ABSTRACT FROM PUBLISHER]

Representing, manipulating, and inferring trust from the user point of view certainly is a grand challenge in virtual worlds, including online games. When someone meets an unknown individual, the question is “Can I trust him/her or not?” This requires the user to have access to a representation of trust about others, as well as a set of operators to undertake inference about the trustability of other users/players. In this paper, we employ a trust representation generated from in-world data in order to feed individual trust decisions. To achieve that purpose, we assume that such a representation of trust already exists; in fact, it was proposed in another paper of ours. Thus, the focus here is on the trust mechanisms required to infer trustability of other users/players. More specifically, we use an individual trust representation deployed as a trust network as base to the inference mechanism that employs two subjective logic operators (consensus and discount) to automatically derive trust decisions. The proposed trust inference system has been validated through OpenSimulator scenarios, which has led to a 5% increase on trustability of avatars in relation to the reference scenario (without trust). [ABSTRACT FROM PUBLISHER]

This paper addresses the problem of automatic analytical pole–zero (PZ) extraction for multistage operational amplifiers (opamps) with frequency compensation. Traditional methods mainly rely on numerical reference to derive approximate PZ expressions without incorporating any design knowledge. Such methods suffer from bad interpretability of the auto-generated results. This paper takes a topological approach and attempts to advocate that certain form of design knowledge can be incorporated in the symbolic term selection process for PZ generation. The generation engine selects the dominant terms by a formal inspection on the token patterns that are correlated to gain factors and compensation elements. Since the gain factors and compensation elements of an opamp are pertinent to the topological details of a circuit, the proposed PZ extraction method is closer to design conception than other numerical reference-based methods. Consequently, the generated pole/zero results are better interpretable. Application to a class of multistage opamps with a variety of compensation structures demonstrates that the proposed method is effective and can match human-derived results. [ABSTRACT FROM PUBLISHER]

Tree reconciliation is a general framework for investigating the mutual influence between gene and species trees according to the parsimony principle, that is, to each evolutionary event a cost is assigned and the goal is to find a reconciliation of minimum total cost. The resulting optimization problem is known as the reconciliation problem. Usually, the considered events are: co-divergence, gene Duplication, horizontal gene Transfer, and gene Loss (DTL model), while in a more conservative setting, gene transfers are not allowed (DL model). The reconciliation problem requires, in the DL model, time linear in the dimension of the two trees and at least quadratic time in the DTL model. Hence, it is reasonable to argue that the introduction of horizontal gene transfers increases the complexity of the problem. Instead, we introduce horizontal gene transfers with some constraints and prove that the problem is still linear in the dimension of the trees. Namely, we allow gene transfers of length bounded by $k=2$ k = 2 , on the basis of the observation that transfers are more likely to occur between closely related species than between distantly related ones. Then we extend the same reasonings to the case in which $k>2$ k > 2 under additional constrains. In this paper we study also another problem related to the reconciliation one, that is optimally rooting one of the two trees when it is not, and also for it we prove similar results. The relevance of this contribution lies in showing that, in the transit from the DL to the DTL model, the computational time does not increase suddenly to quadratic but remains linear in the case when gene transfers are very short (i.e., happening between very close genes). [ABSTRACT FROM AUTHOR]

Grounding natural language in images, such as localizing “the black dog on the left of the tree”, is one of the core problems in artificial intelligence, as it needs to comprehend the fine-grained language compositions. However, existing solutions merely rely on the association between the holistic language features and visual features, while neglect the nature of composite reasoning implied in the language. In this paper, we propose a natural language grounding model that can automatically compose a binary tree structure for parsing the language and then perform visual reasoning along the tree in a bottom-up fashion. We call our model RvG-Tree: Recursive Grounding Tree, which is inspired by the intuition that any language expression can be recursively decomposed into two constituent parts, and the grounding confidence score can be recursively accumulated by calculating their grounding scores returned by the two sub-trees. RvG-Tree can be trained end-to-end by using the Straight-Through Gumbel-Softmax estimator that allows the gradients from the continuous score functions passing through the discrete tree construction. Experiments on several benchmarks show that our model achieves the state-of-the-art performance with more explainable reasoning. [ABSTRACT FROM AUTHOR]

The articles in this special issue focus on cognitive dynamic systems and applications for their use. [ABSTRACT FROM PUBLISHER]