Hatim Chahdi, Nistor Grozavu, I. Mougenot, Laure Berti-Équille, Younès Bennani
Recent studies have shown that the use of a priori knowledge can significantly improve the results of unsupervised classification. However, capturing and formatting such knowledge as constraints is not only very expensive requiring the sustained involvement of an expert but it is also very difficult because some valuable information can be lost when it cannot be encoded as constraints. In this paper, we propose a new constraint-based clustering approach based on ontology reasoning for automatically generating constraints and bridging the semantic gap in satellite image labeling. The use of ontology as a priori knowledge has many advantages that we leverage in the context of satellite image interpretation. The experiments we conduct have shown that our proposed approach can deal with incomplete knowledge while completely exploiting the available one.
{"title":"On the Use of Ontology as A Priori Knowledge into Constrained Clustering","authors":"Hatim Chahdi, Nistor Grozavu, I. Mougenot, Laure Berti-Équille, Younès Bennani","doi":"10.1109/DSAA.2016.72","DOIUrl":"https://doi.org/10.1109/DSAA.2016.72","url":null,"abstract":"Recent studies have shown that the use of a priori knowledge can significantly improve the results of unsupervised classification. However, capturing and formatting such knowledge as constraints is not only very expensive requiring the sustained involvement of an expert but it is also very difficult because some valuable information can be lost when it cannot be encoded as constraints. In this paper, we propose a new constraint-based clustering approach based on ontology reasoning for automatically generating constraints and bridging the semantic gap in satellite image labeling. The use of ontology as a priori knowledge has many advantages that we leverage in the context of satellite image interpretation. The experiments we conduct have shown that our proposed approach can deal with incomplete knowledge while completely exploiting the available one.","PeriodicalId":193885,"journal":{"name":"2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA)","volume":"132 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124511817","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Directed acyclic graphs (DAGs) are commonly used to represent causal relations within a large number of random variables. Estimating DAGs from observational data is a difficult task, it is often impossible to uniquely determine edge direction. The skeleton of the graph, where directions are removed from edges, is often estimated instead. We consider the task of estimating the skeleton of a potentially high-dimensional DAG consisting of Gaussian random variables. A drawback of existing methods is that a prohibitively large number of conditional independence relations need to be tested for. By exploiting properties of common random graph families, we develop a new algorithm that requires conditioning only on small sets of variables. By extending previous theoretical results for undirected graphs to the setting of directed graphs, we prove the consistency of our algorithm, and demonstrate improvements over the state-of-the-art alternative in low and high-dimensional simulation settings. We conclude by applying our proposed algorithm on a real gene expression data set.
{"title":"Causal Structure Learning with Reduced Partial Correlation Thresholding","authors":"A. Sondhi, A. Shojaie","doi":"10.1109/DSAA.2016.68","DOIUrl":"https://doi.org/10.1109/DSAA.2016.68","url":null,"abstract":"Directed acyclic graphs (DAGs) are commonly used to represent causal relations within a large number of random variables. Estimating DAGs from observational data is a difficult task, it is often impossible to uniquely determine edge direction. The skeleton of the graph, where directions are removed from edges, is often estimated instead. We consider the task of estimating the skeleton of a potentially high-dimensional DAG consisting of Gaussian random variables. A drawback of existing methods is that a prohibitively large number of conditional independence relations need to be tested for. By exploiting properties of common random graph families, we develop a new algorithm that requires conditioning only on small sets of variables. By extending previous theoretical results for undirected graphs to the setting of directed graphs, we prove the consistency of our algorithm, and demonstrate improvements over the state-of-the-art alternative in low and high-dimensional simulation settings. We conclude by applying our proposed algorithm on a real gene expression data set.","PeriodicalId":193885,"journal":{"name":"2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA)","volume":"16 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121507959","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Information Retrieval (IR) effectiveness measures commonly assume that the experimental query sets consist of randomly drawn queries that represent the population of queries submitted to IR systems. In many practical situations, however, this assumption is violated, in a problem known as sample selection bias. It follows that the systems participating in evaluation campaigns are ranked by biased estimators of effectiveness. In this paper, we address the problem of query sample selection bias in machine learning terms and study experimentally how retrieval system rankings are affected by it. To this end, we apply a number of retrieval effectiveness measures and query probability estimation methods useful to correct sample selection bias. We report that the ranking of the most effective systems and that of the least effective systems is fairly affected by query sample selection bias, while the ranking of the average systems is much more affected. We also report that the measure of bias depends on the retrieval measure used to rank systems and eventually on the search task being evaluated.
{"title":"Impact of Query Sample Selection Bias on Information Retrieval System Ranking","authors":"M. Melucci","doi":"10.1109/DSAA.2016.43","DOIUrl":"https://doi.org/10.1109/DSAA.2016.43","url":null,"abstract":"Information Retrieval (IR) effectiveness measures commonly assume that the experimental query sets consist of randomly drawn queries that represent the population of queries submitted to IR systems. In many practical situations, however, this assumption is violated, in a problem known as sample selection bias. It follows that the systems participating in evaluation campaigns are ranked by biased estimators of effectiveness. In this paper, we address the problem of query sample selection bias in machine learning terms and study experimentally how retrieval system rankings are affected by it. To this end, we apply a number of retrieval effectiveness measures and query probability estimation methods useful to correct sample selection bias. We report that the ranking of the most effective systems and that of the least effective systems is fairly affected by query sample selection bias, while the ranking of the average systems is much more affected. We also report that the measure of bias depends on the retrieval measure used to rank systems and eventually on the search task being evaluated.","PeriodicalId":193885,"journal":{"name":"2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA)","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122733011","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We study a variation of the classical multi-armed bandits problem. In this problem, the learner has to make a sequence of decisions, picking from a fixed set of choices. In each round, she receives as feedback only the loss incurred from the chosen action. Conventionally, this problem has been studied when losses of the actions are drawn from an unknown distribution or when they are adversarial. In this paper, we study this problem when the losses of the actions also satisfy certain structural properties, and especially, do show a trend structure. When this is true, we show that using trend detection, we can achieve regret of order Õ (N √TK) with respect to a switching strategy for the version of the problem where a single action is chosen in each round and Õ (Nm √TK) when m actions are chosen each round. This guarantee is a significant improvement over the conventional benchmark. Our approach can, as a framework, be applied in combination with various well-known bandit algorithms, like Exp3. For both versions of the problem, we give regret guarantees also for the anytime setting, i.e. when length of the choice-sequence is not known in advance. Finally, we pinpoint the advantages of our method by comparing it to some well-known other strategies.
{"title":"Trend Detection Based Regret Minimization for Bandit Problems","authors":"Paresh Nakhe, Rebecca Reiffenhäuser","doi":"10.1109/DSAA.2016.35","DOIUrl":"https://doi.org/10.1109/DSAA.2016.35","url":null,"abstract":"We study a variation of the classical multi-armed bandits problem. In this problem, the learner has to make a sequence of decisions, picking from a fixed set of choices. In each round, she receives as feedback only the loss incurred from the chosen action. Conventionally, this problem has been studied when losses of the actions are drawn from an unknown distribution or when they are adversarial. In this paper, we study this problem when the losses of the actions also satisfy certain structural properties, and especially, do show a trend structure. When this is true, we show that using trend detection, we can achieve regret of order Õ (N √TK) with respect to a switching strategy for the version of the problem where a single action is chosen in each round and Õ (Nm √TK) when m actions are chosen each round. This guarantee is a significant improvement over the conventional benchmark. Our approach can, as a framework, be applied in combination with various well-known bandit algorithms, like Exp3. For both versions of the problem, we give regret guarantees also for the anytime setting, i.e. when length of the choice-sequence is not known in advance. Finally, we pinpoint the advantages of our method by comparing it to some well-known other strategies.","PeriodicalId":193885,"journal":{"name":"2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA)","volume":"112 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124125434","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Quadratic compressive sensing, as a nonlinear extension of compressive sensing, has attracted considerable attention in optical image, X-ray crystallography, transmission electron microscopy, etc. We introduce the concept of uniform s-regularity to study the uniqueness in quadratic compressive sensing and propose a greedy algorithm for the corresponding numerical optimization. Moreover, we prove the convergence of the proposed algorithm under the uniform s-regularity condition. Finally, we present numerical results to demonstrate the efficiency of the proposed method.
{"title":"The Uniqueness and Greedy Method for Quadratic Compressive Sensing","authors":"Jun Fan, Lingchen Kong, Liqun Wang, N. Xiu","doi":"10.1109/DSAA.2016.94","DOIUrl":"https://doi.org/10.1109/DSAA.2016.94","url":null,"abstract":"Quadratic compressive sensing, as a nonlinear extension of compressive sensing, has attracted considerable attention in optical image, X-ray crystallography, transmission electron microscopy, etc. We introduce the concept of uniform s-regularity to study the uniqueness in quadratic compressive sensing and propose a greedy algorithm for the corresponding numerical optimization. Moreover, we prove the convergence of the proposed algorithm under the uniform s-regularity condition. Finally, we present numerical results to demonstrate the efficiency of the proposed method.","PeriodicalId":193885,"journal":{"name":"2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA)","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128032662","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The Cusp Catastrophe Model provides a promising approach for health and behavioral researchers to investigate both continuous and quantum changes in one modeling framework. However, application of the model is hindered by unresolved issues around a statistical model fitting to the data. This paper reports our exploratory work in developing a new approach to statistical cusp catastrophe modeling. In this new approach, the Cusp Catastrophe Model is cast into a statistical nonlinear regression for parameter estimation. The algorithms of the delayed convention and Maxwell convention are applied to obtain parameter estimates using maximum likelihood estimation. Through a series of simulation studies, we demonstrate that (a) parameter estimation of this statistical cusp model is unbiased, and (b) use of a bootstrapping procedure enables efficient statistical inference. To test the utility of this new method, we analyze survey data collected for an NIH-funded project providing HIV-prevention education to adolescents in the Bahamas. We found that the results can be more reasonably explained by our approach than other existing methods. Additional research is needed to establish this new approach as the most reliable method for fitting the cusp catastrophe model. Further research should focus on additional theoretical analysis, extension of the model for analyzing categorical and counting data, and additional applications in analyzing different data types.
{"title":"An Exploratory Statistical Cusp Catastrophe Model","authors":"D. Chen, X. Chen, Kai Zhang","doi":"10.1109/DSAA.2016.17","DOIUrl":"https://doi.org/10.1109/DSAA.2016.17","url":null,"abstract":"The Cusp Catastrophe Model provides a promising approach for health and behavioral researchers to investigate both continuous and quantum changes in one modeling framework. However, application of the model is hindered by unresolved issues around a statistical model fitting to the data. This paper reports our exploratory work in developing a new approach to statistical cusp catastrophe modeling. In this new approach, the Cusp Catastrophe Model is cast into a statistical nonlinear regression for parameter estimation. The algorithms of the delayed convention and Maxwell convention are applied to obtain parameter estimates using maximum likelihood estimation. Through a series of simulation studies, we demonstrate that (a) parameter estimation of this statistical cusp model is unbiased, and (b) use of a bootstrapping procedure enables efficient statistical inference. To test the utility of this new method, we analyze survey data collected for an NIH-funded project providing HIV-prevention education to adolescents in the Bahamas. We found that the results can be more reasonably explained by our approach than other existing methods. Additional research is needed to establish this new approach as the most reliable method for fitting the cusp catastrophe model. Further research should focus on additional theoretical analysis, extension of the model for analyzing categorical and counting data, and additional applications in analyzing different data types.","PeriodicalId":193885,"journal":{"name":"2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130345474","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
T. Guo, Zhao Xu, X. Yao, Hai-Ming Chen, K. Aberer, K. Funaya
Time series forecasting for streaming data plays an important role in many real applications, ranging from IoT systems, cyber-networks, to industrial systems and healthcare. However the real data is often complicated with anomalies and change points, which can lead the learned models deviating from the underlying patterns of the time series, especially in the context of online learning mode. In this paper we present an adaptive gradient learning method for recurrent neural networks (RNN) to forecast streaming time series in the presence of anomalies and change points. We explore the local features of time series to automatically weight the gradients of the loss of the newly available observations with distributional properties of the data in real time. We perform extensive experimental analysis on both synthetic and real datasets to evaluate the performance of the proposed method.
{"title":"Robust Online Time Series Prediction with Recurrent Neural Networks","authors":"T. Guo, Zhao Xu, X. Yao, Hai-Ming Chen, K. Aberer, K. Funaya","doi":"10.1109/DSAA.2016.92","DOIUrl":"https://doi.org/10.1109/DSAA.2016.92","url":null,"abstract":"Time series forecasting for streaming data plays an important role in many real applications, ranging from IoT systems, cyber-networks, to industrial systems and healthcare. However the real data is often complicated with anomalies and change points, which can lead the learned models deviating from the underlying patterns of the time series, especially in the context of online learning mode. In this paper we present an adaptive gradient learning method for recurrent neural networks (RNN) to forecast streaming time series in the presence of anomalies and change points. We explore the local features of time series to automatically weight the gradients of the loss of the newly available observations with distributional properties of the data in real time. We perform extensive experimental analysis on both synthetic and real datasets to evaluate the performance of the proposed method.","PeriodicalId":193885,"journal":{"name":"2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA)","volume":"206 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133365515","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Vincent Etter, M. E. Khan, M. Grossglauser, Patrick Thiran
We consider online predictions of vote results, where regions across a country vote on an issue under discussion. Such online predictions before and during the day of the vote are useful to media agencies, polling institutes, and political parties, e.g., to identify regions that are crucial in determining the national outcome of a vote. We analyze a unique dataset from Switzerland. The dataset contains 281 votes from 2352 regions over a period of 34 years. We make several contributions towards improving online predictions. First, we show that these votes exhibit a bi-clustering of the vote results, i.e., regions that are spatially close tend to vote similarly, and issues that discuss similar topics show similar global voting patterns. Second, we develop models that can exploit this bi-clustering, as well as the features associated with the votes and regions. Third, we show that, when combining vote results and features together, Bayesian methods are essential to obtaining good performance. Our results show that Bayesian methods give better estimates of the hyperparameters than non-Bayesian methods such as cross-validation. The resulting models generalize well to many different tasks, produce robust predictions, and are easily interpretable.
{"title":"Online Collaborative Prediction of Regional Vote Results","authors":"Vincent Etter, M. E. Khan, M. Grossglauser, Patrick Thiran","doi":"10.1109/DSAA.2016.31","DOIUrl":"https://doi.org/10.1109/DSAA.2016.31","url":null,"abstract":"We consider online predictions of vote results, where regions across a country vote on an issue under discussion. Such online predictions before and during the day of the vote are useful to media agencies, polling institutes, and political parties, e.g., to identify regions that are crucial in determining the national outcome of a vote. We analyze a unique dataset from Switzerland. The dataset contains 281 votes from 2352 regions over a period of 34 years. We make several contributions towards improving online predictions. First, we show that these votes exhibit a bi-clustering of the vote results, i.e., regions that are spatially close tend to vote similarly, and issues that discuss similar topics show similar global voting patterns. Second, we develop models that can exploit this bi-clustering, as well as the features associated with the votes and regions. Third, we show that, when combining vote results and features together, Bayesian methods are essential to obtaining good performance. Our results show that Bayesian methods give better estimates of the hyperparameters than non-Bayesian methods such as cross-validation. The resulting models generalize well to many different tasks, produce robust predictions, and are easily interpretable.","PeriodicalId":193885,"journal":{"name":"2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA)","volume":"83 S366","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132226896","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Martin Wistuba, Nicolas Schilling, L. Schmidt-Thieme
Algorithm selection and hyperparameter tuning are omnipresent problems for researchers and practitioners. Hence, it is not surprising that the efforts in automatizing this process using various meta-learning approaches have been increased. Sequential model-based optimization (SMBO) is ne of the most popular frameworks for finding optimal hyperparameter configurations. Originally designed for black-box optimization, researchers have contributed different meta-learning approaches to speed up the optimization process. We create a generalized framework of SMBO and its recent additions which gives access to adaptive hyperparameter transfer learning with simple surrogates (AHT), a new class of hyperparameter optimization strategies. AHT provides less time-overhead for the optimization process by replacing time-and space-consuming transfer surrogate models with simple surrogates that employ adaptive transfer learning. In an empirical comparison on two different meta-data sets, we can show that AHT outperforms various instances of the SMBO framework in the scenarios of hyperparameter tuning and algorithm selection.
{"title":"Hyperparameter Optimization Machines","authors":"Martin Wistuba, Nicolas Schilling, L. Schmidt-Thieme","doi":"10.1109/DSAA.2016.12","DOIUrl":"https://doi.org/10.1109/DSAA.2016.12","url":null,"abstract":"Algorithm selection and hyperparameter tuning are omnipresent problems for researchers and practitioners. Hence, it is not surprising that the efforts in automatizing this process using various meta-learning approaches have been increased. Sequential model-based optimization (SMBO) is ne of the most popular frameworks for finding optimal hyperparameter configurations. Originally designed for black-box optimization, researchers have contributed different meta-learning approaches to speed up the optimization process. We create a generalized framework of SMBO and its recent additions which gives access to adaptive hyperparameter transfer learning with simple surrogates (AHT), a new class of hyperparameter optimization strategies. AHT provides less time-overhead for the optimization process by replacing time-and space-consuming transfer surrogate models with simple surrogates that employ adaptive transfer learning. In an empirical comparison on two different meta-data sets, we can show that AHT outperforms various instances of the SMBO framework in the scenarios of hyperparameter tuning and algorithm selection.","PeriodicalId":193885,"journal":{"name":"2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA)","volume":"244 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114440240","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Many classification problems involve nodes that have a natural connection between them, such as links between people, pages, or social network accounts. Recent work has demonstrated how to learn relational dependencies from these links, then leverage them as predictive features. However, while this can often improve accuracy, the use of linked information can also lead to cascading prediction errors, especially in the common-case when a network is only sparsely-labeled. In response, this paper examines several existing and new methods for correcting the "relational bias" that leads to such errors. First, we explain how existing approaches can be divided into "resemblance-based" and "assignment-based" methods, and provide the first experimental comparison between them. We demonstrate that all of these methods can improve accuracy, but that the former type typically leads to better accuracy. Moreover, we show that the more flexible methods typically perform best, motivating a new assignment-based method that often improves accuracy vs. a more rigid method. In addition, we demonstrate for the first time that some of these methods can also improve accuracy when combined with Gibbs sampling for inference. However, we show that, with Gibbs, correcting relational bias also requires improving label initialization, and present two new initialization methods that yield large accuracy gains. Finally, we evaluate the effects of relational bias when "neighbor attributes," recently-proposed additions that can provide more stability during inference, are included as model features. We show that such attributes reduce the negative impact of bias, but that using some form of bias correction remains important for achieving maximal accuracy.
{"title":"Correcting Relational Bias to Improve Classification in Sparsely-Labeled Networks","authors":"J. R. King, Luke K. McDowell","doi":"10.1109/DSAA.2016.11","DOIUrl":"https://doi.org/10.1109/DSAA.2016.11","url":null,"abstract":"Many classification problems involve nodes that have a natural connection between them, such as links between people, pages, or social network accounts. Recent work has demonstrated how to learn relational dependencies from these links, then leverage them as predictive features. However, while this can often improve accuracy, the use of linked information can also lead to cascading prediction errors, especially in the common-case when a network is only sparsely-labeled. In response, this paper examines several existing and new methods for correcting the \"relational bias\" that leads to such errors. First, we explain how existing approaches can be divided into \"resemblance-based\" and \"assignment-based\" methods, and provide the first experimental comparison between them. We demonstrate that all of these methods can improve accuracy, but that the former type typically leads to better accuracy. Moreover, we show that the more flexible methods typically perform best, motivating a new assignment-based method that often improves accuracy vs. a more rigid method. In addition, we demonstrate for the first time that some of these methods can also improve accuracy when combined with Gibbs sampling for inference. However, we show that, with Gibbs, correcting relational bias also requires improving label initialization, and present two new initialization methods that yield large accuracy gains. Finally, we evaluate the effects of relational bias when \"neighbor attributes,\" recently-proposed additions that can provide more stability during inference, are included as model features. We show that such attributes reduce the negative impact of bias, but that using some form of bias correction remains important for achieving maximal accuracy.","PeriodicalId":193885,"journal":{"name":"2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115770200","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：