On one hand, compared with traditional rela- tional and XML models, graphs have more expressive power and are widely used today. On the other hand, various ap- plications of social computing trigger the pressing need ...On one hand, compared with traditional rela- tional and XML models, graphs have more expressive power and are widely used today. On the other hand, various ap- plications of social computing trigger the pressing need of a new search paradigm. In this article, we argue that big graph search is the one filling this gap. We first introduce the ap- plication of graph search in various scenarios. We then for- malize the graph search problem, and give an analysis of graph search from an evolutionary point of view, followed by the evidences from both the industry and academia. After that, we analyze the difficulties and challenges of big graph search. Finally, we present three classes of techniques to- wards big graph search： query techniques, data techniques and distributed computing techniques.展开更多
An iterative procedure introduced in MacKay’s evidence framework is often used for estimating the hyperparameter in empirical Bayes.Together with the use of a particular form of prior,the estimation of the hyperparam...An iterative procedure introduced in MacKay’s evidence framework is often used for estimating the hyperparameter in empirical Bayes.Together with the use of a particular form of prior,the estimation of the hyperparameter reduces to an automatic relevance determination model,which provides a soft way of pruning model parameters.Despite the effectiveness of this estimation procedure,it has stayed primarily as a heuristic to date and its application to deep neural network has not yet been explored.This paper formally investigates the mathematical nature of this procedure and justifies it as a well-principled algorithm framework,which we call the MacKay algorithm.As an application,we demonstrate its use in deep neural networks,which have typically complicated structure with millions of parameters and can be pruned to reduce the memory requirement and boost computational efficiency.In experiments,we adopt MacKay algorithm to prune the parameters of both simple networks such as LeNet,deep convolution VGG-like networks,and residual netowrks for large image classification task.Experimental results show that the algorithm can compress neural networks to a high level of sparsity with little loss of prediction accuracy,which is comparable with the state-of-the-art.展开更多
The increasing availability of folksonomy data makes them vital for user profiling approaches to precisely detect user preferences and better understand user interests, so as to render some personalized recommendation...The increasing availability of folksonomy data makes them vital for user profiling approaches to precisely detect user preferences and better understand user interests, so as to render some personalized recommendation or re- trieval results. This paper presents a rigorous probabilis- tic framework to discover user preference from folkson- omy data. Furthermore, we incorporate three models into the framework with the corresponding inference methods, expectation-maximization or Gibbs sampling algorithms. The user preference is expressed through topical conditional distributions. Moreover, to demonstrate the versatility of our framework, a recommendation method is introduced to show the possible usage of our framework and evaluate the applica- bility of the engaged models. The experimental results show that, with the help of the proposed framework, the user pref- erence can be effectively discovered.展开更多
基金This work was supported in part by 973 program （2014CB340300）, National Natural Science Foundation of China （Grant No. 61322207） and the Fundamental Research Funds for the Central Universi- ties.
文摘On one hand, compared with traditional rela- tional and XML models, graphs have more expressive power and are widely used today. On the other hand, various ap- plications of social computing trigger the pressing need of a new search paradigm. In this article, we argue that big graph search is the one filling this gap. We first introduce the ap- plication of graph search in various scenarios. We then for- malize the graph search problem, and give an analysis of graph search from an evolutionary point of view, followed by the evidences from both the industry and academia. After that, we analyze the difficulties and challenges of big graph search. Finally, we present three classes of techniques to- wards big graph search： query techniques, data techniques and distributed computing techniques.
基金This work was supported partly by China Scholarship Council(201706020062)by China 973 program(2015CB358700)+2 种基金by the National Natural Science Foundation of China(Grant Nos.61772059,61421003)by the Beijing Advanced Innovation Center for Big Data and Brain Computing(BDBC)State Key Laboratory of Software Development Environment(SKLSDE-2018ZX-17).
文摘An iterative procedure introduced in MacKay’s evidence framework is often used for estimating the hyperparameter in empirical Bayes.Together with the use of a particular form of prior,the estimation of the hyperparameter reduces to an automatic relevance determination model,which provides a soft way of pruning model parameters.Despite the effectiveness of this estimation procedure,it has stayed primarily as a heuristic to date and its application to deep neural network has not yet been explored.This paper formally investigates the mathematical nature of this procedure and justifies it as a well-principled algorithm framework,which we call the MacKay algorithm.As an application,we demonstrate its use in deep neural networks,which have typically complicated structure with millions of parameters and can be pruned to reduce the memory requirement and boost computational efficiency.In experiments,we adopt MacKay algorithm to prune the parameters of both simple networks such as LeNet,deep convolution VGG-like networks,and residual netowrks for large image classification task.Experimental results show that the algorithm can compress neural networks to a high level of sparsity with little loss of prediction accuracy,which is comparable with the state-of-the-art.
基金This work was supported by the National Basic Re-search program of China （2014CB340305）, partly by the National Natural Science Foundation of China （Grant Nos. 61300070 and 61421003） and partly by the State Key Lab for Software Development Environment.
文摘The increasing availability of folksonomy data makes them vital for user profiling approaches to precisely detect user preferences and better understand user interests, so as to render some personalized recommendation or re- trieval results. This paper presents a rigorous probabilis- tic framework to discover user preference from folkson- omy data. Furthermore, we incorporate three models into the framework with the corresponding inference methods, expectation-maximization or Gibbs sampling algorithms. The user preference is expressed through topical conditional distributions. Moreover, to demonstrate the versatility of our framework, a recommendation method is introduced to show the possible usage of our framework and evaluate the applica- bility of the engaged models. The experimental results show that, with the help of the proposed framework, the user pref- erence can be effectively discovered.
