In order to increase the autonomy of the modern, high complexity robots with multiple degrees of freedom, it is necessary for them to be able to learn and adapt their skills, for example, using reinforcement learning ...
详细信息
In order to increase the autonomy of the modern, high complexity robots with multiple degrees of freedom, it is necessary for them to be able to learn and adapt their skills, for example, using reinforcement learning (RL). However, RL performance greatly depends on the task dimensionality. Methods for reducing the task dimensionality, such as deep autoencoder neural networks, are often employed. Such neural network based dimensionality reduction approaches require a large example database for training, but obtaining such a database for a real robot is a complex and tedious process. This paper proposes a method of obtaining a database for the training of a deep autoencoder network, which serves for the dimensionality reduction of robot learning, and thus accelerates the robot's ability to adapt to the real world. The presented method is based on a few real-world examples and statistical generalization. A comparison to using a simulated-only database on the use-case of robot throwing shows that the proposed approach achieves better real-world performance.
This paper presents a novel method to reconstruct and separate metabolite and macromolecule (MM) signals in H-1 magnetic resonance spectroscopic imaging (MRSI) data using learned nonlinear models. Specifically, deep a...
详细信息
ISBN:
(纸本)9781538693308
This paper presents a novel method to reconstruct and separate metabolite and macromolecule (MM) signals in H-1 magnetic resonance spectroscopic imaging (MRSI) data using learned nonlinear models. Specifically, deep autoencoder (DAE) networks were constructed and trained to learn the nonlinear low-dimensional manifolds, where the metabolite and MM signals reside individually. A regularized reconstruction formulation is proposed to integrate the learned models with signal encoding model to reconstruct and separate the metabolite and MM components. An efficient algorithm was developed to solve the associated optimization problem. The performance of the proposed method has been evaluated using simulation and experimental H-1-MRSI data. Efficient low-dimensional signal representation of the learned models and improved metabolite/MM separation over the standard parametric fitting based approach have been demonstrated.
Recommendation systems have been used widely in many industries, including online retail, movies, and news media. Indeed, video game recommendation systems are one of the most important tools available to users and ga...
详细信息
ISBN:
(纸本)9781450371056
Recommendation systems have been used widely in many industries, including online retail, movies, and news media. Indeed, video game recommendation systems are one of the most important tools available to users and game distribution platforms today. A good recommender can help customers find games they might like faster, not only making it easier for them but also helping game distributors and developers to increase their sales and improve customer satisfaction ratings. One such platform that can greatly benefit from a recommendation system is Steam, the largest digital PC game distribution platform. Steam sees over a dozen million users login every day. It collects a considerable amount of data on each user, and this data may be used to help make better game recommendations. This paper proposes STEAMer, a solution for a new video game recommendation system for the Steam platform. STEAMer utilizes the Steam user data in conjunction with a deep autoencoder learning model to generate potential recommendations;we also apply the additional user data to an existing deep neural network-based recommendation system. Performance evaluation shows that the additional user data does indeed improve recommendation performance. Furthermore, when both systems use the additional user data, the deep autoencoder-based STEAMer still proves superior to the baseline deep neural network-based system in both mean average precision @ 10 (MAP@10) and normalized discounted cumulative gain @ 10 (NDCG@10) scores and in diversity.
To enhance features of different electromagnetic interference (EMI) signals, which are significant for further feature extraction and pattern recognition, the authors propose an EMI signal feature enhancement method b...
详细信息
To enhance features of different electromagnetic interference (EMI) signals, which are significant for further feature extraction and pattern recognition, the authors propose an EMI signal feature enhancement method based on extreme energy difference and a deep auto-encoder. Experimental results show that this method can effectively enhance features of EMI signals and improve recognition accuracy.
Attributed network embedding is the task to learn a lower dimensional vector representation of the nodes of an attributed network, which can be used further for downstream network mining tasks. Nodes in a network exhi...
详细信息
ISBN:
(纸本)9781450368223
Attributed network embedding is the task to learn a lower dimensional vector representation of the nodes of an attributed network, which can be used further for downstream network mining tasks. Nodes in a network exhibit community structure and most of the network embedding algorithms work well when the nodes, along with their attributes, adhere to the community structure of the network. But real life networks come with community outlier nodes, which deviate significantly in terms of their link structure or attribute similarities from the other nodes of the community they belong to. These outlier nodes, if not processed carefully, can even affect the embeddings of the other nodes in the network. Thus, a node embedding framework for dealing with both the link structure and attributes in the presence of outliers in an unsupervised setting is practically important. In this work, we propose a deep unsupervised autoencoders based solution which minimizes the effect of outlier nodes while generating the network embedding. We use both stochastic gradient descent and closed form updates for faster optimization of the network parameters. We further explore the role of adversarial learning for this task, and propose a second unsupervised deep model which learns by discriminating the structure and the attribute based embeddings of the network and minimizes the effect of outliers in a coupled way. Our experiments show the merit of these deep models to detect outliers and also the superiority of the generated network embeddings for different downstream mining tasks. To the best of our knowledge, these are the first unsupervised non linear approaches that reduce the effect of the outlier nodes while generating Network Embedding.
Community detection is a classic and challenging network analysis task. Inspired by the similarity between autoencoder and modularity maximization model in terms of a low-dimensional approximation of the modularity ma...
详细信息
ISBN:
(纸本)9781728196190
Community detection is a classic and challenging network analysis task. Inspired by the similarity between autoencoder and modularity maximization model in terms of a low-dimensional approximation of the modularity matrix. In this paper, a novel community detection framework via deep autoencoder was proposed. Firstly, the modularity is fed into the autoencoder, which can obtain a non-linear deep representation for the network. In addition, node attribute similarity was utilized to construct pairwise constraints on nodes, and then a graph regularization is introduced into the framework. The extensive experimental evaluations on real-world networks and synthetic networks demonstrate that the proposed framework achieves superior performance compared with the state-of-the-art community detection algorithms.
Multi-view subspace clustering aims to find the inherent structure of data as much as possible by fusing complementary information of multiple views to achieve better clustering ***,most of the traditional multiview s...
详细信息
Multi-view subspace clustering aims to find the inherent structure of data as much as possible by fusing complementary information of multiple views to achieve better clustering ***,most of the traditional multiview subspace clustering algorithms are only shallow clustering algorithms,which does not capture the deep information of the data well,and does not conduct in-depth research at the self-representation level of the *** this end,this paper proposes a novel deep multi-view subspace clustering model that introduces exclusive constraints.A deep autoencoder is used to perform nonlinear low-dimensional subspace mapping for each view to learn the deep structure of the original *** better retain multiple views’ local structure and better reflect the complementarity,the exclusive constraints are introduced into the self-representation matrix which located in the middle layer of the deep *** multi-view consensus self-representation matrix is used to capture the consistency information between the multi-view *** update of autoencoder parameters and clustering parameters are iteratively optimized under the same learning framework to improve the clustering *** on multi-view data sets prove that this method can better dig out the inherent complementary structure of multi-view data,which reflects the superiority of this method.
The aim of manifold learning is to extract low-dimensional manifolds from high-dimensional data. Manifold alignment is a variant of manifold learning that uses two or more datasets that are assumed to represent differ...
详细信息
The aim of manifold learning is to extract low-dimensional manifolds from high-dimensional data. Manifold alignment is a variant of manifold learning that uses two or more datasets that are assumed to represent different high-dimensional representations of the same underlying manifold. Manifold alignment can be successful in detecting latent manifolds in cases where one version of the data alone is not sufficient to extract and establish a stable low-dimensional representation. The present study proposes a parallel deep autoencoder neural network architecture for manifold alignment and conducts a series of experiments using a protein-folding benchmark dataset and a suite of new datasets generated by simulating double-pendulum dynamics with underlying manifolds of dimensions 2, 3 and 4. The dimensionality and topological complexity of these latent manifolds are above those occurring in most previous studies. Our experimental results demonstrate that the parallel deep autoencoder performs in most cases better than the tested traditional methods of semi-supervised manifold alignment. We also show that the parallel deep autoencoder can process datasets of different input domains by aligning the manifolds extracted from kinematics parameters with those obtained from corresponding image data.
Soil liquefaction assessment remains a crucial and complex challenge in seismic geotechnical engineering due to various liquefaction records and limited information, which entails a more generalized off-the-shelf mode...
详细信息
Soil liquefaction assessment remains a crucial and complex challenge in seismic geotechnical engineering due to various liquefaction records and limited information, which entails a more generalized off-the-shelf model that can achieve favourable performance on different data sources. In this work, a deep learning model is built and investigated on the soil liquefaction prediction and a modified transfer learning scheme between different data sources is presented. Various datasets, including shear wave velocity-based, CPT-based, SPT-based, and real cases, are collected and utilized to verify the effectiveness and accuracy of the proposed model. Because different data sources in soil liquefaction generally share several geotechnical and mechanical parameters, we work to combine model prior information, feature mapping and data reconstruction in transfer learning models to tackle multi-source domain adaption, which can be further applied to other predictive analysis and facilitate online learning models in geotechnical engineering. Also, the deep learning model is compared with several classical machine learning and ensemble learning models and the modified transfer learning model is formulated by comparing different feature transformation techniques integrated with various feature-based and instance-based transfer learning methods. The accuracy and effectiveness of the deep learning and modified transfer learning models have been validated in the numerical study.
deep learning has a strong ability to extract feature representations from data, since it has a great advantage in processing nonlinear and non-stationary data and reflecting nonlinear interactive relationship. This p...
详细信息
deep learning has a strong ability to extract feature representations from data, since it has a great advantage in processing nonlinear and non-stationary data and reflecting nonlinear interactive relationship. This paper proposes to apply deep learning algorithms including deep neural network and deep autoencoder to track index performance and introduces a dynamic weight calculation method to measure the direct effects of the stocks on index. The empirical study takes historical data of Hang Seng Index (HSI) and its constituents to analyze the effectiveness and practicability of the index tracking method. The results show that the index tracking method based on deep neural network has a smaller tracking error, and thus can effectively track the index. (C) 2018 Elsevier B.V. All rights reserved.
暂无评论