摘要 :
A crucial assumption in most statistical learning theory is that samples are independently and identically distributed (i.i.d.). However, for many real applications, the i.i.d. assumption does not hold. We consider learning proble...
展开
A crucial assumption in most statistical learning theory is that samples are independently and identically distributed (i.i.d.). However, for many real applications, the i.i.d. assumption does not hold. We consider learning problems in which examples are dependent and their dependency relation is characterized by a graph. To establish algorithm-dependent generalization theory for learning with non-i.i.d. data, we first prove novel McDiarmid-type concentration inequalities for Lipschitz functions of graph-dependent random variables. We show that concentration relies on the forest complexity of the graph, which characterizes the strength of the dependency. We demonstrate that for many types of dependent data, the forest complexity is small and thus implies good concentration. Based on our new inequalities, we establish stability bounds for learning graph-dependent data.
收起
摘要 :
With the rapid development of Artificial Intelligence (AI) in the Internet of things (IoT), information fusion becomes one of the key technology. Serial fusion plays an important role in information fusion, which concatenates the ...
展开
With the rapid development of Artificial Intelligence (AI) in the Internet of things (IoT), information fusion becomes one of the key technology. Serial fusion plays an important role in information fusion, which concatenates the features obtained from different sources to get a super-feature with higher discrimination. However, the performance of the serial fusion is usually affected by the poor quality of the indistinguishable real-world datasets, especially the problem of noise and imbalance. The existing noise data removal methods may destroy the data distribution, and the existing oversampling methods may create noise samples. They may affect the result of feature reduction and further reduce the performance of the classification. To address these big practical challenges, a novel Small-Groups-based Serial Fusion method (SGSF) is proposed. SGSF consists of two main parts: a Small-Groups-based Noise Removal method (SGNR) and a small-groups-based Positive Region Oriented Oversampling method (PROO). Both of these two methods are based on the Neighborhood Rough Set (NRS) theory. SGNR identifies and removes low-density noise samples whilst keeps the high-density outlier samples, targeting to enlarge the limited NRS positive region and preserve the data distribution. PROO generates new samples towards the NRS positive region so that it can balance the dataset and increase the number of samples with more explicit features. Intensive experiments are carried out to verify the performance of SGSF based on 3 publicly available datasets and 6 self-collected medical datasets. The results demonstrate that SGSF increases the classification accuracy from existing 82.2% to 94.3%.
收起
摘要 :
With the rapid development of Artificial Intelligence (AI) in the Internet of things (IoT), information fusion becomes one of the key technology. Serial fusion plays an important role in information fusion, which concatenates the ...
展开
With the rapid development of Artificial Intelligence (AI) in the Internet of things (IoT), information fusion becomes one of the key technology. Serial fusion plays an important role in information fusion, which concatenates the features obtained from different sources to get a super-feature with higher discrimination. However, the performance of the serial fusion is usually affected by the poor quality of the indistinguishable real-world datasets, especially the problem of noise and imbalance. The existing noise data removal methods may destroy the data distribution, and the existing oversampling methods may create noise samples. They may affect the result of feature reduction and further reduce the performance of the classification. To address these big practical challenges, a novel Small-Groups-based Serial Fusion method (SGSF) is proposed. SGSF consists of two main parts: a Small-Groups-based Noise Removal method (SGNR) and a small-groups-based Positive Region Oriented Oversampling method (PROO). Both of these two methods are based on the Neighborhood Rough Set (NRS) theory. SGNR identifies and removes low-density noise samples whilst keeps the high-density outlier samples, targeting to enlarge the limited NRS positive region and preserve the data distribution. PROO generates new samples towards the NRS positive region so that it can balance the dataset and increase the number of samples with more explicit features. Intensive experiments are carried out to verify the performance of SGSF based on 3 publicly available datasets and 6 self-collected medical datasets. The results demonstrate that SGSF increases the classification accuracy from existing 82.2% to 94.3%.
收起
摘要 :
Image paragraph captioning (IPC) aims to generate a fine-grained paragraph to describe the visual content of an image. Significant progress has been made by deep neural networks, in which the attention mechanism plays an essential...
展开
Image paragraph captioning (IPC) aims to generate a fine-grained paragraph to describe the visual content of an image. Significant progress has been made by deep neural networks, in which the attention mechanism plays an essential role. However, conventional attention mechanisms tend to ignore the past alignment information, which often results in problems of repetitive captioning and incomplete captioning. In this paper, we propose an Interactive key-value Memory-augmented Attention model for image Paragraph captioning (IMAP) to keep track of the attention history (salient objects coverage information) along with the update-chain of the decoder state and therefore avoid generating repetitive or incomplete image descriptions. In addition, we employ an adaptive attention mechanism to realize adaptive alignment from image regions to caption words, where an image region can be mapped to an arbitrary number of caption words while a caption word can also attend to an arbitrary number of image regions. Extensive experiments on a benchmark dataset (i.e., Stanford) demonstrate the effectiveness of our IMAP model.
收起
摘要 :
Image paragraph captioning (IPC) aims to generate a fine-grained paragraph to describe the visual content of an image. Significant progress has been made by deep neural networks, in which the attention mechanism plays an essential...
展开
Image paragraph captioning (IPC) aims to generate a fine-grained paragraph to describe the visual content of an image. Significant progress has been made by deep neural networks, in which the attention mechanism plays an essential role. However, conventional attention mechanisms tend to ignore the past alignment information, which often results in problems of repetitive captioning and incomplete captioning. In this paper, we propose an Interactive key-value Memory-augmented Attention model for image Paragraph captioning (IMAP) to keep track of the attention history (salient objects coverage information) along with the update-chain of the decoder state and therefore avoid generating repetitive or incomplete image descriptions. In addition, we employ an adaptive attention mechanism to realize adaptive alignment from image regions to caption words, where an image region can be mapped to an arbitrary number of caption words while a caption word can also attend to an arbitrary number of image regions. Extensive experiments on a benchmark dataset (i.e., Stanford) demonstrate the effectiveness of our IMAP model.
收起
摘要 :
Behavior Tree, known as modular and reactive, is widely used in industry and research. Its variant, the Reactive Behavior Tree (RBT), can effectively resolve the unpredictable environments through dynamic expansion based on action...
展开
Behavior Tree, known as modular and reactive, is widely used in industry and research. Its variant, the Reactive Behavior Tree (RBT), can effectively resolve the unpredictable environments through dynamic expansion based on actions and predicates. However, in partially observable environments, considering the observe limitation of scene, the existing RBT expanding method will break the necessary sequential dependence between the predicates and cause the agent to fail to perceive state changes in the environment, leaving the planner in a spin state. We proposed an optimized RBT method, where the RBT can repair the sequential dependence by rollbacking dynamically. We verified the optimized method in a simulation environment, and our method was shown to improve the effectiveness and robustness of RBT in partially observable environments.
收起
摘要 :
Service response time is considered as one of the most important quality parameters of a web service. This paper focuses on the microservices selection problem that how to select a service invocation path with the shortest respons...
展开
Service response time is considered as one of the most important quality parameters of a web service. This paper focuses on the microservices selection problem that how to select a service invocation path with the shortest response time. We propose a novel distributed microservices selection framework (DMSF) which is distributed and can be easily integrated into common microservices architectures. A distributed microservices selection algorithm (DMSA) is developed within the framework, which incorporates the aspects of both source routing algorithm and distributed routing algorithm. DMSF attempts to reduce the time complexity of service selection and communication overhead while ensuring the shortest response time. The experiment results show that DMSF outperforms the existing frameworks in terms of time complexity, communication overhead and response time.
收起
原文获取