arXiv subjects

Zhiguang Wang

Publications and source records attributed to Zhiguang Wang.

27 records · Page 2Linked to original sources

arXiv2016

Time Series Classification from Scratch with Deep Neural Networks: A Strong Baseline

We propose a simple but strong baseline for time series classification from scratch with deep neural networks. Our proposed baseline models are pure end-to-end without any heavy preprocessing on the raw data or feature crafting. The proposed Fully Convolutional Network (FCN) achieves premium performance to other state-of-the-art approaches and our exploration of the very deep neural networks with the ResNet structure is also competitive. The global average pooling in our convolutional model enables the exploitation of the Class Activation Map (CAM) to find out the contributing region in the raw data for the specific labels. Our models provides a simple choice for the real world application and a good starting point for the future research. An overall analysis is provided to discuss the generalization capability of our models, learned features, network structures and the classification semantics.

cs.LG↗

arXiv2016

Representation Learning with Deconvolution for Multivariate Time Series Classification and Visualization

We propose a new model based on the deconvolutional networks and SAX discretization to learn the representation for multivariate time series. Deconvolutional networks fully exploit the advantage the powerful expressiveness of deep neural networks in the manner of unsupervised learning. We design a network structure specifically to capture the cross-channel correlation with deconvolution, forcing the pooling operation to perform the dimension reduction along each position in the individual channel. Discretization based on Symbolic Aggregate Approximation is applied on the feature vectors to further extract the bag of features. We show how this representation and bag of features helps on classification. A full comparison with the sequence distance based approach is provided to demonstrate the effectiveness of our approach on the standard datasets. We further build the Markov matrix from the discretized representation from the deconvolution to visualize the time series as complex networks, which show more class-specific statistical properties and clear structures with respect to different labels.

cs.LG↗

arXiv2016

Adaptive Normalized Risk-Averting Training For Deep Neural Networks

This paper proposes a set of new error criteria and learning approaches, Adaptive Normalized Risk-Averting Training (ANRAT), to attack the non-convex optimization problem in training deep neural networks (DNNs). Theoretically, we demonstrate its effectiveness on global and local convexity lower-bounded by the standard $L_p$-norm error. By analyzing the gradient on the convexity index $λ$, we explain the reason why to learn $λ$ adaptively using gradient descent works. In practice, we show how this method improves training of deep neural networks to solve visual recognition tasks on the MNIST and CIFAR-10 datasets. Without using pretraining or other tricks, we obtain results comparable or superior to those reported in recent literature on the same tasks using standard ConvNets + MSE/cross entropy. Performance on deep/shallow multilayer perceptrons and Denoised Auto-encoders is also explored. ANRAT can be combined with other quasi-Newton training methods, innovative network variants, regularization techniques and other specific tricks in DNNs. Other than unsupervised pretraining, it provides a new perspective to address the non-convex optimization problem in DNNs.

cs.LG↗

arXiv2015

Spatially Encoding Temporal Correlations to Classify Temporal Data Using Convolutional Neural Networks

We propose an off-line approach to explicitly encode temporal patterns spatially as different types of images, namely, Gramian Angular Fields and Markov Transition Fields. This enables the use of techniques from computer vision for feature learning and classification. We used Tiled Convolutional Neural Networks to learn high-level features from individual GAF, MTF, and GAF-MTF images on 12 benchmark time series datasets and two real spatial-temporal trajectory datasets. The classification results of our approach are competitive with state-of-the-art approaches on both types of data. An analysis of the features and weights learned by the CNNs explains why the approach works.

cs.LG↗

arXiv2015

Empirical Studies on Symbolic Aggregation Approximation Under Statistical Perspectives for Knowledge Discovery in Time Series

Symbolic Aggregation approXimation (SAX) has been the de facto standard representation methods for knowledge discovery in time series on a number of tasks and applications. So far, very little work has been done in empirically investigating the intrinsic properties and statistical mechanics in SAX words. In this paper, we applied several statistical measurements and proposed a new statistical measurement, i.e. information embedding cost (IEC) to analyze the statistical behaviors of the symbolic dynamics. Our experiments on the benchmark datasets and the clinical signals demonstrate that SAX can always reduce the complexity while preserving the core information embedded in the original time series with significant embedding efficiency. Our proposed IEC score provide a priori to determine if SAX is adequate for specific dataset, which can be generalized to evaluate other symbolic representations. Our work provides an analytical framework with several statistical tools to analyze, evaluate and further improve the symbolic dynamics for knowledge discovery in time series.

cs.LG↗

arXiv2015

Imaging Time-Series to Improve Classification and Imputation

Inspired by recent successes of deep learning in computer vision, we propose a novel framework for encoding time series as different types of images, namely, Gramian Angular Summation/Difference Fields (GASF/GADF) and Markov Transition Fields (MTF). This enables the use of techniques from computer vision for time series classification and imputation. We used Tiled Convolutional Neural Networks (tiled CNNs) on 20 standard datasets to learn high-level features from the individual and compound GASF-GADF-MTF images. Our approaches achieve highly competitive results when compared to nine of the current best time series classification approaches. Inspired by the bijection property of GASF on 0/1 rescaled data, we train Denoised Auto-encoders (DA) on the GASF images of four standard and one synthesized compound dataset. The imputation MSE on test data is reduced by 12.18%-48.02% when compared to using the raw data. An analysis of the features and weights learned via tiled CNNs and DAs explains why the approaches work.

cs.LG↗

arXiv2015

Self-blocking of interstitial clusters near metallic grain boundaries

Nano-crystallize materials have been known for decades to potentially owe the novel self-healing ability for radiation damage, which has been demonstrated to be especially linked to preferential occupation of interstitials at grain boundary (GB) and promoted vacancy-interstitial annihilation. A major obstacle to better understanding the healing property is the lack of an atomistic picture of the interstitial states near GBs, due to severely separation of the timescale of interstitial segregation from other events and abundance of interstitials at the GB. Here, we report a generic "self-blocking" effect of the interstitial cluster (SIAn) near the metallic GB in W, Mo and Fe. Upon creating a SIAn near the GB, it is immediately trapped by the GB during the GB structural relaxation and blocks there, impeding GB's further spontaneous trapping of the SIAn in the vicinity and making these SIAns stuck nearby the GB. The SIAn in the stuck state surprisingly owes an exceptionally larger number of annihilation sites with vacancies near the GB than the SIAn trapped at the GB due to maintaining its bulk configuration basically. Besides, it also has an unexpectedly long-ranged repelling interaction with the SIA in the bulk region, which may further affect the GB's trap of the SIAn. The self-blocking effect might shed light on more critical and extended role of the GB in healing radiation-damage in NCs than previously recognized the GB's limited role based on GB's trap for the SIA and resulted vacancy-SIA recombination.

cond-mat.mtrl-sci↗

arXiv2014

Crafting the strain state in epitaxial thin films: A case study of CoFe2O4 films on Pb(Mg,Nb)O3-PbTiO3

The strain dependence of electric and magnetic properties has been widely investigated, both from a fundamental science perspective and an applications point of view. Electromechanical coupling through field-induced polarization rotation (PRO) and polarization reorientation (PRE) in piezoelectric single crystals can provide an effective strain in film/substrate epitaxial heterostructures. However, the specific pathway of PRO and PRE is a complex thermodynamic process, depending on chemical composition, temperature, electric field and mechanical load. Here, systematic studies of the temperature-dependent field-induced phase transitions in Pb(Mg,Nb)O3-PbTiO3 single crystals with different initial phase and orientation configurations have been performed. Different types of strains, volatile/nonvolatile and biaxial/uniaxial, have been measured by both macroscopic and in-situ X-ray diffraction techniques. In addition, the strain state of epitaxial Mn-doped CoFe2O4 thin films was examined by magnetic anisotropy measurements, where a giant magnetoelectric coupling has been demonstrated.

cond-mat.mtrl-sci↗

arXiv2013

An operational window for radiation-resistant materials based on sequentially healing grain interiors and boundaries

Design of nuclear materials with high radiation-tolerance has great significance1, especially for the next generation of nuclear energy systems2,3. Response of nano- and poly-crystals to irradiation depends on the radiation temperature, dose-rate and grain size4-13. However the dependencies had been studied and interpreted individually, and thus severely lacking is the ability to predict radiation performance of materials in extreme environments. Here we propose an operational window for radiation-resistant materials, which is based on a perspective of interactions among irradiation-induced interstitials, vacancies, and grain boundaries. Using atomic simulations, we find that healing grain boundaries needs much longer time than healing grain interiors. Not been noticed before, this finding suggests priority should be thereafter given to recovery of the grain boundary itself. This large disparity in healing time is reflected in the spectra of defects-recombination energy barriers by the presence of one high-barrier peak in addition to the peak of low barriers. The insight gained from the study instigates new avenues for examining the role of grain boundaries in healing the material. In particular, we sketch out the radiation-endurance window in the parameter space of temperature, dose-rate and grain size. The window helps evaluate material performance and develop resistant materials against radiation damage.

cond-mat.mtrl-sci↗