arXiv subjects

Yue Fan

Publications and source records attributed to Yue Fan.

63 records · Page 4Linked to original sources

arXiv2020

Structural origin of plasticity in strained high-entropy alloy

High-entropy alloys (HEAs) are solid solutions of multiple elements with equal atomic ratios which present an innovative pathway for de novo alloy engineering. While there exist extensive studies to ascertain the important structural aspects governing their mechanical behaviors, elucidating the underlying deformation mechanisms still remains a challenge. Using atomistic simulations, we probe the particle rearrangements in a yielding, model HEA system to understand the structural origin of its plasticity. We find the plastic deformation is initiated by irreversible topological fluctuations which tend to spatially localize in regions termed as soft spots which consist of particles actively participating in slow vibrational motions, an observation strikingly reminiscent of nonlinear glassy rheology. Due to the varying local elastic moduli resulting from the loss of compositional periodicity, these plastic responses exhibit significant spatial heterogeneity and are found to be inversely correlated with the distribution of local electronegativity. Further mechanical loading promotes the cooperativity among these local plastic events and triggers the formation of dislocation loops. As in strained crystalline solids, different dislocation loops can further merge together and propagate as the main carrier of large-scale plastic deformation. However, the energy barriers located at the spatial regions with higher local electronegativity severely hinders the motion of dislocations. By delineating the transient mechanical response in terms of atomic configuration, our computational findings shed new light on understanding the nature of plasticity of single-phase HEA.

cond-mat.mtrl-sci↗

arXiv2020

Construction of the Moduli Space of Higgs Bundles using Analytic Methods

It is a folklore theorem that the Kuranishi slice method can be used to construct the moduli space of semistable Higgs bundles on a closed Riemann surface as a complex space. The purpose of this paper is to provide a proof in detail. We also give a direct proof that the moduli space is locally modeled on an affine GIT quotient of a quadratic cone by a complex reductive group.

math.DG↗

arXiv2020

Soft mode parameter as an indicator for the activation energy spectra in metallic glass

The activation energy (E_A) spectra of potential energy landscape (PEL) provides a convenient perspective for interpreting complex phenomena in amorphous materials; however, the link between the E_A spectra and other physical properties in metallic glasses is still mysterious. By systematically probing the E_A spectra for numerous metallic glass samples with distinct local geometric ordering, which correspond to broad processing histories, it is found that the shear modulus of the samples are strongly correlated with the arithmetic mean of the E_A spectra rather than with the local geometrical ordering. Furthermore, we studied the correlation of the obtained E_A spectra and various well-established physical parameters. The outcome of our research clearly demonstrates that the soft mode parameter Ψ and the E_A spectrum are correlated; therefore, it could be a good indicator of metallic glass properties and sheds important light on the structure-property relationship in metallic glass through the medium of PEL.

cond-mat.mtrl-sci↗

arXiv2020

Analyzing the Dependency of ConvNets on Spatial Information

Intuitively, image classification should profit from using spatial information. Recent work, however, suggests that this might be overrated in standard CNNs. In this paper, we are pushing the envelope and aim to further investigate the reliance on spatial information. We propose spatial shuffling and GAP+FC to destroy spatial information during both training and testing phases. Interestingly, we observe that spatial information can be deleted from later layers with small performance drops, which indicates spatial information at later layers is not necessary for good performance. For example, test accuracy of VGG-16 only drops by 0.03% and 2.66% with spatial information completely removed from the last 30% and 53% layers on CIFAR100, respectively. Evaluation on several object recognition datasets (CIFAR100, Small-ImageNet, ImageNet) with a wide range of CNN architectures (VGG16, ResNet50, ResNet152) shows an overall consistent pattern.

cs.CV↗

arXiv2019

CN-CELEB: a challenging Chinese speaker recognition dataset

Recently, researchers set an ambitious goal of conducting speaker recognition in unconstrained conditions where the variations on ambient, channel and emotion could be arbitrary. However, most publicly available datasets are collected under constrained environments, i.e., with little noise and limited channel variation. These datasets tend to deliver over optimistic performance and do not meet the request of research on speaker recognition in unconstrained conditions. In this paper, we present CN-Celeb, a large-scale speaker recognition dataset collected `in the wild'. This dataset contains more than 130,000 utterances from 1,000 Chinese celebrities, and covers 11 different genres in real world. Experiments conducted with two state-of-the-art speaker recognition approaches (i-vector and x-vector) show that the performance on CN-Celeb is far inferior to the one obtained on VoxCeleb, a widely used speaker recognition dataset. This result demonstrates that in real-life conditions, the performance of existing techniques might be much worse than it was thought. Our database is free for researchers and can be downloaded from http://project.cslt.org.

eess.AS↗

arXiv2018

Parameter-Free Spatial Attention Network for Person Re-Identification

Global average pooling (GAP) allows to localize discriminative information for recognition [40]. While GAP helps the convolution neural network to attend to the most discriminative features of an object, it may suffer if that information is missing e.g. due to camera viewpoint changes. To circumvent this issue, we argue that it is advantageous to attend to the global configuration of the object by modeling spatial relations among high-level features. We propose a novel architecture for Person Re-Identification, based on a novel parameter-free spatial attention layer introducing spatial relations among the feature map activations back to the model. Our spatial attention layer consistently improves the performance over the model without it. Results on four benchmarks demonstrate a superiority of our model over the state-of-the-art achieving rank-1 accuracy of 94.7% on Market-1501, 89.0% on DukeMTMC-ReID, 74.9% on CUHK03-labeled and 69.7% on CUHK03-detected.

cs.CV↗

arXiv2018

Transcription Factor-DNA Binding Via Machine Learning Ensembles

We present ensemble methods in a machine learning (ML) framework combining predictions from five known motif/binding site exploration algorithms. For a given TF the ensemble starts with position weight matrices (PWM's) for the motif, collected from the component algorithms. Using dimension reduction, we identify significant PWM-based subspaces for analysis. Within each subspace a machine classifier is built for identifying the TF's gene (promoter) targets (Problem 1). These PWM-based subspaces form an ML-based sequence analysis tool. Problem 2 (finding binding motifs) is solved by agglomerating k-mer (string) feature PWM-based subspaces that stand out in identifying gene targets. We approach Problem 3 (binding sites) with a novel machine learning approach that uses promoter string features and ML importance scores in a classification algorithm locating binding sites across the genome. For target gene identification this method improves performance (measured by the F1 score) by about 10 percentage points over the (a) motif scanning method and (b) the coexpression-based association method. Top motif outperformed 5 component algorithms as well as two other common algorithms (BEST and DEME). For identifying individual binding sites on a benchmark cross species database (Tompa et al., 2005) we match the best performer without much human intervention. It also improved the performance on mammalian TFs. The ensemble can integrate orthogonal information from different weak learners (potentially using entirely different types of features) into a machine learner that can perform consistently better for more TFs. The TF gene target identification component (problem 1 above) is useful in constructing a transcriptional regulatory network from known TF-target associations. The ensemble is easily extendable to include more tools as well as future PWM-based information.

q-bio.GN↗

arXiv2013

Feature vector regularization in machine learning

Problems in machine learning (ML) can involve noisy input data, and ML classification methods have reached limiting accuracies when based on standard ML data sets consisting of feature vectors and their classes. Greater accuracy will require incorporation of prior structural information on data into learning. We study methods to regularize feature vectors (unsupervised regularization methods), analogous to supervised regularization for estimating functions in ML. We study regularization (denoising) of ML feature vectors using Tikhonov and other regularization methods for functions on ${\bf R}^n$. A feature vector ${\bf x}=(x_1,\ldots,x_n)=\{x_q\}_{q=1}^n$ is viewed as a function of its index $q$, and smoothed using prior information on its structure. This can involve a penalty functional on feature vectors analogous to those in statistical learning, or use of proximity (e.g. graph) structure on the set of indices. Such feature vector regularization inherits a property from function denoising on ${\bf R}^n$, in that accuracy is non-monotonic in the denoising (regularization) parameter $α$. Under some assumptions about the noise level and the data structure, we show that the best reconstruction accuracy also occurs at a finite positive $α$ in index spaces with graph structures. We adapt two standard function denoising methods used on ${\bf R}^n$, local averaging and kernel regression. In general the index space can be any discrete set with a notion of proximity, e.g. a metric space, a subset of ${\bf R}^n$, or a graph/network, with feature vectors as functions with some notion of continuity. We show this improves feature vector recovery, and thus the subsequent classification or regression done on them. We give an example in gene expression analysis for cancer classification with the genome as an index space and network structure based protein-protein interactions.

stat.ML↗

arXiv2013

MetSizeR: selecting the optimal sample size for metabolomic studies using an analysis based approach

Background: Determining sample sizes for metabolomic experiments is important but due to the complexity of these experiments, there are currently no standard methods for sample size estimation in metabolomics. Since pilot studies are rarely done in metabolomics, currently existing sample size estimation approaches which rely on pilot data can not be applied. Results: In this article, an analysis based approach called MetSizeR is developed to estimate sample size for metabolomic experiments even when experimental pilot data are not available. The key motivation for MetSizeR is that it considers the type of analysis the researcher intends to use for data analysis when estimating sample size. MetSizeR uses information about the data analysis technique and prior expert knowledge of the metabolomic experiment to simulate pilot data from a statistical model. Permutation based techniques are then applied to the simulated pilot data to estimate the required sample size. Conclusions: The MetSizeR methodology, and a publicly available software package which implements the approach, are illustrated through real metabolomic applications. Sample size estimates, informed by the intended statistical analysis technique, and the associated uncertainty are provided.

stat.AP↗