Different Levels Of Granularity Research Articles

Rough possibilistic C-means clustering based on multigranulation approximation regions and shadowed sets

The management of uncertain information in a data set is crucial for clustering models. In this study, we present a rough possibilistic C-means clustering approach based on multigranulation approximation regions and shadowed sets, which can handle the uncertainties implicated in data and generated by the model parameters simultaneously. In particular, all patterns are first partitioned into three approximation regions with respect to a fixed cluster according to their possibilistic membership degrees based on shadowed set theory, which can help capture the natural topology of the data, especially when dealing with outliers and noisy data. The multigranulation approximation regions of each cluster can then be formed under a series of fuzzifier values, where the uncertainty caused by a specific fuzzifier value can be detected based on variations in the approximation regions with different levels of granularity. We also introduce a framework for updating prototypes based on ensemble strategies to attenuate the distortions due to iteration during clustering procedures. Finally, an adaptive mechanism is developed for dynamically adjusting scale parameters based on the notion of the maximal compatible regions of clusters. By integrating various granular computing techniques, i.e., rough sets, fuzzy sets, shadowed sets, and the notion of multigranulation, the uncertainties implicated in data and produced by model parameters can be adequately addressed, and the possibilistic membership values involved make the method sufficiently robust to deal with noisy environments. The improved performance of the proposed approach was demonstrated in experiments based on the comparisons with other available fuzzy and possibilistic clustering methods.

Snorkel MeTaL: Weak Supervision for Multi-Task Learning.

Many real-world machine learning problems are challenging to tackle for two reasons: (i) they involve multiple sub-tasks at different levels of granularity; and (ii) they require large volumes of labeled training data. We propose Snorkel MeTaL, an end-to-end system for multi-task learning that leverages weak supervision provided at multiple levels of granularity by domain expert users. In MeTaL, a user specifies a problem consisting of multiple, hierarchically-related sub-tasks-for example, classifying a document at multiple levels of granularity-and then provides labeling functions for each sub-task as weak supervision. MeTaL learns a re-weighted model of these labeling functions, and uses the combined signal to train a hierarchical multi-task network which is automatically compiled from the structure of the sub-tasks. Using MeTaL on a radiology report triage task and a fine-grained news classification task, we achieve average gains of 11.2 accuracy points over a baseline supervised approach and 9.5 accuracy points over the predictions of the user-provided labeling functions.

2251 Assessing research impact: It takes a team

OBJECTIVES/SPECIFIC AIMS: Dissemination of research findings through the published literature is a complex but critical part of the scholarly communication process. Additionally, this time point on the translational spectrum is a key objective of the National Clinical Association for Advancing Translational Sciences (NCATS). Tracking the dissemination of research outputs can be difficult to identify and evaluate. The purpose of this case study was 2-fold: (1) identify tools and resources available freely to the public and through university subscriptions used to assess research output; and (2) compare the effectiveness of these tools oat tracking output at different levels of granularity. METHODS/STUDY POPULATION: The authors, Spectrum staff (D.A.) and School of Medicine librarian (M.B.), attended webinars hosted by other Academic Medical Center libraries conducting work on impact tracking and learned from vendor product managers about available tools and resources during on-site campus visits. Publications from Stanford’s Clinical and Translational Science Award (CTSA) were used to track the diffusion of research outputs (e.g., number of citations, document types, research areas, relative citation ratio, CTSAs collaboration) via library subscription services (e.g., Web of Science and Scopus) and freely available tools (e.g., iCite and PubMed). RESULTS/ANTICIPATED RESULTS: The authors found certain tools were more inclusive in retrieving grant funded research outputs. For example, in the case of UL1 grant (UL1TR001085, UL1TR000093, UL1RR025744), on a grant-level output, there were discrepancies in the number of publications retrieved: (1) PubMed found 644 outputs; (2) Web of Science found 497 outputs; and (3) Scopus found 190 outputs. After de-duplication, the search across Web of Science (WoS), Scopus, and PubMed yielded 899 publications. In total, 389 outputs were unique to PubMed; 165 were unique to WoS; and 90 were unique to Scopus. Future analysis will be conducted to identify the source of unique outputs from each database (e.g., conference proceeding, specific journals). Additional analysis based on other units of research outputs (e.g., author-level outputs and article-level outputs) are expected to yield similar discrepancies. DISCUSSION/SIGNIFICANCE OF IMPACT: Citation analysis is a valuable method of assessing research output and, to a larger extent, research impact in a given field. It can help investigators illustrate qualifications for undertaking new projects, highlight collaborations across schools and departments, justify a grant renewal, and/or highlight accomplishments for promotion. However, systematic and comprehensive evaluations are needed in tandem with citation analysis/bibliometric analysis to assess the translation and uptake of research outputs and activities that result in research impact. Furthermore, both investigators and staff need adequate time and training to process research outputs/activities and to effectively organize them in easily understood visualizations.

Multigranulation rough-fuzzy clustering based on shadowed sets

In this study, a new technique of rough-fuzzy clustering based on multigranulation approximation regions is developed to tackle the uncertainty associated with the fuzzifier parameter m. According to shadowed set theory, the multigranulation approximation regions for each cluster can be formed based on fuzzy membership degrees under the multiple values of fuzzifier parameter with a partially ordered relation. The uncertainty generated by the fuzzifier parameter m can be captured and interpreted through the variations in approximation regions among different levels of granularity, rather than at a single level of granularity under a specific fuzzifier value. An ensemble strategy for updating prototypes is then presented based on the constructed multigranulation approximation regions, in which the prototype calculations that may be spoiled due to the uncertainty caused by a single fuzzifier value can be modified. Finally, a multilevel degranulation mechanism is introduced to evaluate the validity of clustering methods. By integrating the notions of shadowed sets and multigranulation into rough-fuzzy clustering approaches, the overall topology of data can be captured well and the uncertain information implicated in data can be effectively addressed, including the uncertainty generated by fuzzification coefficient, the vagueness arising in boundary regions and overlapping partitions. The essence of the proposed method is illustrated by comparative experiments in terms of several validity indices.

Decomposition of the uncovered equity parity correlation

Central to the uncovered equity parity (UEP) condition are the sign and magnitude of the observed UEP correlation. In this paper, we decompose the UEP correlation at two different levels of granularity: a macro (bilateral) level and a micro (multilateral) level. We achieve the decomposition of the UEP correlation by first decomposing the local price of each local equity market into two global (multilateral) prices: the global price of the local equity market; and the global price of the local currency. We find that the heterogeneity of the local currencies provides a number of explanations for the empirical inconsistency of both the sign and the magnitude of the observed UEP correlations.

The role of baseline granularity for benchmarking citation impact. The case of CSS profiles

In this paper we study the effect of granularity on Characteristic Scores and Scales (CSS). Unlike the traditional indicators that are mostly based on means and quantiles, CSS require the reduction of the citation distributions collaboration of the underlying reference population to four states (classes) and thus higher a different level of granularity. While the question of the choice of granularity is at higher levels of aggregation usually not critical since countries and university have rather multidisciplinary profiles, at lower aggregation levels specialisation becomes more typical. Inappropriate granularity might not warrant the depiction of the publication profiles at these levels in a correct and adequate manner and thus not add accurate citation profiles either. In order to be able to process one complete annual volume of the Web of Science, we decided to calculate CSS thresholds and classes for two levels of granularity, namely sub-fields and WoS Subject Categories. With about 5% deviation, we did not find a real significance. However, we identified journals with similar impact measures but different citation profiles, independently of the granularity. Finally, we have pointed to the limitations in the choice of granularity—in terms of both too broad and too narrow subjects.

Assertion-Based QA With Question-Aware Open Information Extraction

We present assertion based question answering (ABQA), an open domain question answering task that takes a question and a passage as inputs, and outputs a semi-structured assertion consisting of a subject, a predicate and a list of arguments. An assertion conveys more evidences than a short answer span in reading comprehension, and it is more concise than a tedious passage in passage-based QA. These advantages make ABQA more suitable for human-computer interaction scenarios such as voice-controlled speakers. Further progress towards improving ABQA requires richer supervised dataset and powerful models of text understanding. To remedy this, we introduce a new dataset called WebAssertions, which includes hand-annotated QA labels for 358,427 assertions in 55,960 web passages. To address ABQA, we develop both generative and extractive approaches. The backbone of our generative approach is sequence to sequence learning. In order to capture the structure of the output assertion, we introduce a hierarchical decoder that first generates the structure of the assertion and then generates the words of each field. The extractive approach is based on learning to rank. Features at different levels of granularity are designed to measure the semantic relevance between a question and an assertion. Experimental results show that our approaches have the ability to infer question-aware assertions from a passage. We further evaluate our approaches by incorporating the ABQA results as additional features in passage-based QA. Results on two datasets show that ABQA features significantly improve the accuracy on passage-based QA.

Developing an automated writing placement system for ESL learners

ABSTRACTThere are quite a few challenges in the development of an automated writing placement model for non-native English learners, among them the fact that exams that encompass the full range of language proficiency exhibited at different stages of learning are hard to design. However, acquisition of appropriate training data that are relevant to the task at hand is essential in the development of the model. Using the Cambridge Learner Corpus writing scores, which have been subsequently benchmarked to Common European Framework of Reference for Languages (CEFR) levels, we conceptualize the task as a supervised machine learning problem, and primarily focus on developing a generic writing model. Such an approach facilitates the modeling of truly consistent, internal marking criteria regardless of the prompt delivered, which has the additional advantage of requiring smaller dataset sizes and not necessarily requiring re-training or tuning for new tasks. The system is developed to predict someone’s proficiency level on the CEFR scale, which allows learners to point to a specific standard of achievement. We furthermore integrate our model into Cambridge English Write & ImproveTM—a freely available, cloud-based tool that automatically provides diagnostic feedback to non-native English language learners at different levels of granularity—and examine its use.

Building Optimal Multiresolution Soft Sensors for Continuous Processes

Data-driven models used in soft sensor applications are expected to capture the dominant relationships between the different process variables and the outputs, while accounting for their high-dimensional, dynamic, and multiresolution character. While the first two characteristics are often addressed, the multiresolution aspect is usually disregarded and confused with a multirate scenario: multiresolution occurs when variables have different levels of granularity, because of, for instance, automatic averaging operations over certain time windows; on the other hand, a multirate structure is caused by the existence of different sampling rates, but the granularity of the recorded values is the same. This has two major and immediate implications. First, current methods are unable to handle variables with different resolutions in a consistent and rigorous way, since they tacitly assume that data represent instant observations and not averages over time windows. Second, even if data is available at a single-resolution (i.e., all variables with the same granularity), it is not guaranteed that the native resolution of the predictors is the most appropriate for modeling. Therefore, soft sensor development must address not only the selection of the best set of predictors to be included in the model, but also the optimum resolution to adopt for each predictor. In this work, two novel multiresolution frameworks for soft sensor development are proposed (MRSS-SC and MRSS-DC) that actively introduce multiresolution into the data by searching for the best granularity for each variable. The performance of these methodologies is comparatively assessed against current single-resolution counterparts. The optimized multiresolution soft sensors are bounded to be at least as good as their single-resolution versions, and the results confirm that they almost always perform substantially better.

Set-theoretical and Combinatorial Instruments for Problem Space Analysis in Adaptive Serious Games

The Computerized Adaptive Practice (CAP) system describes a set of algorithms for assessing player’s expertise and difficulties of in-game problems and for adapting the latter to the former. However, an effective use of CAP requires that in-game problems are designed carefully and refined over time to avoid possible barriers to learning. This study proposes a methodology and three different instruments for analyzing the problem set in CAP-enabled games. The instruments include the Guttman scale, a ranked order, and a Hasse diagram that offer analysis at different levels of granularity and complexity. The methodology proposes to use quantified difficulty measures to infer topology of the problem set. It is well-suited for serious games that emphasize practice and repetitive play. The emphasis is put on the simplicity of use and visualization of the problem space to maximally support teachers and game developers in designing and refining CAP-enabled games. Two case studies demonstrate practical applications of the proposed instruments on empirical data. Future research directions are proposed to address potential drawbacks.

Learning multi-grained aspect target sequence for Chinese sentiment analysis

Aspect-based sentiment analysis aims at identifying sentiment polarity towards aspect targets in a sentence. Previously, the task was modeled as a sentence-level sentiment classification problem that treated aspect targets as a hint. Such approaches oversimplify the problem by averaging word embeddings when the aspect target is a multi-word sequence. In this paper, we formalize the problem from a different perspective, i.e., that sentiment at aspect target level should be the main focus. Due to the fact that written Chinese is very rich and complex, Chinese aspect targets can be studied at three different levels of granularity: radical, character and word. Thus, we propose to explicitly model the aspect target and conduct sentiment classification directly at the aspect target level via three granularities. Moreover, we study two fusion methods for such granularities in the task of Chinese aspect-level sentiment analysis. Experimental results on a multi-word aspect target subset from SemEval2014 and four Chinese review datasets validate our claims and show the improved performance of our model over the state of the art.

Multi-granularity unbalanced linguistic group decision-making with incomplete weight information based on VIKOR method

Decision makers (DMs) adopt multi-granularity linguistic assessments to form decision matrixes as a result of different backgrounds and experience, which is common in the process of group decision-making (GDM). Especially, unbalanced linguistic term sets (ULTS), as a representative linguistic representation, are employed in numerous fields. To deal with the problem of different granularities, the linguistic hierarchies (LH) are introduced to transform different level of granularities into the same level. In addition, incomplete attribute weights are considered, which are obtained by solving a multi-objective optimization model. Furthermore, extending Vlsekriterijumska Optimizacija I Kompromisno Resenje (VIKOR) method to multi-granularity ULTS with incomplete attribute weight in GDM is proposed. Finally, an example concerning green supplier selection and a comparative analysis are carried out to illustrate the proposed method and find that group or personal opinions can be reflected according to actual situations compared with other approaches.

Convolutional adaptive denoising autoencoders for hierarchical feature extraction

Convolutional neural networks (CNNs) are typical structures for deep learning and are widely used in image recognition and classification. However, the random initialization strategy tends to become stuck at local plateaus or even diverge, which results in rather unstable and ineffective solutions in real applications. To address this limitation, we propose a hybrid deep learning CNN-AdapDAE model, which applies the features learned by the AdapDAE algorithm to initialize CNN filters and then train the improved CNN for classification tasks. In this model, AdapDAE is proposed as a CNN pre-training procedure, which adaptively obtains the noise level based on the principle of annealing, by starting with a high level of noise and lowering it as the training progresses. Thus, the features learned by AdapDAE include a combination of features at different levels of granularity. Extensive experimental results on STL-10, CIFAR-10, andMNIST datasets demonstrate that the proposed algorithm performs favorably compared to CNN (random filters), CNNAE (pre-training filters by autoencoder), and a few other unsupervised feature learning methods.

Multi-Scale Attentive Interaction Networks for Chinese Medical Question Answer Selection

The past few years have witnessed a trend that the deep learning techniques have been increasingly applied in healthcare due to the explosive growth of big data. The online medical community, where users can ask qualified doctors about medical questions with just a few keystrokes and mouse clicks anytime and anywhere, has become quite popular recently. In this paper, we investigate the problem of Chinese medical question answer selection, which is a crucial subtask of automatic question answering and fairly challenging because of its language and domain characteristics. We introduce an end-to-end multiscale interactive networks framework to address the issue. The framework consists of several multi-scale deep neural layers which extract the deep semantic information of medical text from different levels of granularity, shortcut connections which prevent network degradation problem, and attentive interaction which mines the correlation between questions and answers. To evaluate our framework, we update and expand a dataset called cMedQA v2.0. Experimental results demonstrate that our model outperforms the existing state-ofthe-art models with noticeable margins.

Models for Interoperable Human Robot Collaboration

In this initial research work, different approaches that allow to represent production processes that can be shared by humans and robots are analyzed. Focus is given to approaches that support modular process model structures, allowing to hide details of how tasks are implemented. This is necessary, as a single task might be realized by a human or robot, and both agents need the task description on a different level of granularity.

Modeling Topic Evolution in Twitter: An Embedding-Based Approach

In last two decades, online social networks have grown vertically as well as horizontally. Due to various users activities in these networks, huge amount of data, mainly textual, is being generated that can be analyzed at different levels of granularity for various purposes, including behavior analysis, sentiment analysis, and predictive modeling. In this paper, we propose a word embedding-based approach to analyze users-centric tweets to observe their behavior evolution in terms of the topics discussed by them over a period of time. We also present a word embedding-based proximity measure to monitor temporal transitions between the topics using five topic evolution events– emergence , persistence , convergence , divergence , and extinction . The proximity between a pair of topics is defined as a function of the content and contextual similarity between their word distributions, wherein the contextual similarity is calculated using word embedding. The proposed approach is evaluated over three Twitter datasets in line with the existing state-of-the-art approaches in literature and the experimental results are encouraging.

Participation, Co-Creation, and Public Space

<p>A central notion in urban design, urban interaction design, and placemaking is the user of public space, the occupant, resident, citizen, bystander, passer-by, explorer, or flâneur. When the field of human-computer interaction (HCI) first emerged, the disciplines that represented the “human” aspects of HCI included behavioural psychology, cognitive science and human factors engineering. This situatedness begs the question whether the “user” requires different contextualisations beyond the immediate and traditional HCI concerns of the technical interface, that is, beyond usability.<br />This article aims to illustrate the need for placemakers and urban interaction designers to be transdisciplinary and agile in order to navigate different levels of granularity. This article seeks to practice granular agile thinking by introducing five possible ways to think about the “urban user” and the implications that follow: the user as city resident; the user as consumer of city services; the user as participant in the city’s community consultations; the user as co-creator in a collaborative approach to citymaking, and finally; the user re-thought as part of a much larger and more complex ecosystem of more-than-human worlds and of cohabitation – a process that decentres the human in the design of collaborative cities.</p>

New hybrid heuristic algorithm for the clustered traveling salesman problem

Abstract In this study, we present a new hybrid heuristic algorithm for the Clustered Traveling Salesman Problem (CTSP). The CTSP is NP -hard because it includes the well-known Traveling Salesman Problem as a special case. In the CTSP, the vertices are partitioned into clusters and all the vertices of each cluster have to be visited contiguously. The algorithm is based on iterated metaheuristics, including local search, greedy randomized adaptive search procedure, and variable neighborhood random descent. Our new hybrid heuristic algorithm uses several variable neighborhood structures, which are selected in a random order. The new algorithm helps to intensify the search by using local search operators and diversification with a constructive heuristic and a perturbation method. The results of computational tests are reported for different classes of instances based on data with different levels of granularity, i.e., with different numbers of vertices and clusters. The results obtained by our new hybrid heuristic were compared with those produced using four previously reported heuristics and an exact method. The computational results demonstrate that the new hybrid heuristic obtained better results when it was compared with a hybrid heuristic that uses variable neighborhood descent. The new hybrid heuristic also outperformed previously described algorithms and it was competitive within a reasonable amount of computational time.

Exploring the relation between referencing practices and citation impact: A large‐scale study based on Web of Science data

In this large‐scale contribution, we deal with the relationship between properties of cited references of Web of Science articles and the field normalized citation rate of these articles. Using nearly 1 million articles, and three classification systems with different levels of granularity, we study the effects of number of cited references, share of references covered by Web of Science, mean age of references and mean citation rate of references on field normalized citation rate. To expose the relationship between the predictor variables and the response variable, we use quantile regression. We found that a higher number of references, a higher share of references to publications within Web of Science and references to more recent publications correlate with citation impact. A correlation was observed even when normalization was done with a finely grained classification system. The predictor variables affected citation impact to a larger extent at higher quantile levels. Regarding the relative importance of the predictor variables, citation impact of the cited references was in general the least important variable. Number of cited references carried most of the importance for both low and medium quantile levels, but this importance was lessened at the highest considered level.

Aggregating Association Rules to Improve Change Recommendation

As the complexity of software systems grows, it becomes increasingly difficult for developers to be aware of all the dependencies that exist between artifacts (e.g., files or methods) of a system. Change recommendation has been proposed as a technique to overcome this problem, as it suggests to a developer relevant source-code artifacts related to her changes. Association rule mining has shown promise in deriving such recommendations by uncovering relevant patterns in the system’s change history. The strength of the mined association rules is captured using a variety of interestingness measures. However, state-of-the-art recommendation engines typically use only the rule with the highest interestingness value when more than one rule applies. In contrast, we argue that when multiple rules apply, this indicates collective evidence, and aggregating those rules (and their evidence) will lead to more accurate change recommendation. To investigate this hypothesis we conduct a large empirical study of 15 open source software systems and two systems from our industry partners. We evaluate association rule aggregation using four variants of the change history for each system studied, enabling us to compare two different levels of granularity in two different scenarios. Furthermore, we study 40 interestingness measures using the rules produced by two different mining algorithms. The results show that (1) between 13 and 90% of change recommendations can be improved by rule aggregation, (2) rule aggregation almost always improves change recommendation for both algorithms and all measures, and (3) fine-grained histories benefit more from rule aggregation.