arXiv subjects

Bo Han

Publications and source records attributed to Bo Han.

At least 361 records · Page 20Linked to original sources

arXiv2020

Hamiltonian approach to the torsional anomalies and its dimensional ladder

Torsion can cause various anomalies in various dimensions, including the $\left(3+1\right)$-dimensional $[(3+1)D]$ Nieh-Yan anomaly, the $\left(2+1\right)$D Hughes-Leigh-Fradkin (HLF) parity anomaly, and the $\left(3+1\right)$D, $\left(1+1\right)$D chiral energy-momentum anomaly. We study these anomalies from the Hamiltonian approach. We derive the $\left(1+1\right)$D chiral energy-momentum anomaly from the single-body Hamiltonian. We then show how other torsional anomalies can be related to the $\left(1+1\right)$D chiral energy-momentum anomaly in a straightforward way. Finally, the Nieh-Yan anomaly and the $\left(3+1\right)$D chiral energy-momentum anomaly are obtained from the parity anomaly and the HLF effective action, respectively. Hence, we have constructed the dimensional ladder for the torsional anomalies from the single-body Hamiltonian picture.

cond-mat.mes-hall↗

arXiv2020

Nieh-Yan Anomaly: Torsional Landau Levels, central charge and anomalous thermal Hall effect

The Nieh-Yan anomaly is the anomalous breakdown of the chiral U(1) symmetry caused by the interaction between torsion and fermions. We study this anomaly from the point of view of torsional Landau levels. It was found that the torsional Landau levels are gapless, while their contributions to the chiral anomaly are canceled, except those from the lowest torsional Landau levels. Hence, the dimension is effectively reduced from (3+1)-dimensional to (1+1)-dimensional. We further show that the coefficient of the Nieh-Yan anomaly is the free energy density in (1+1) dimensions. Especially, at finite temperature, the thermal Nieh-Yan anomaly is proportional to the central charge. The anomalous thermal Hall conductance in Weyl semimetals is then shown to be proportional to the central charge, which is the experimental fingerprint of the thermal Nieh-Yan anomaly.

cond-mat.mes-hall↗

arXiv2020

Torsional Anomalies and Bulk-Dislocation Correspondence in Weyl Systems

Based on the supersymmetric quantum mechanical approach, we have systematically studied both the $U\left(1\right)$ gauge anomaly and the diffeomorphism anomaly in Weyl systems with torsion, curvature and external electromagnetic fields. These anomalies relate to the chiral current (or current) non-conservation and chiral energy-momentum (or energy-momentum) non-conservation, respectively, which can be applied to the $^{3}\text{He-A}$ phase, the chiral superconductors and the Weyl semimetals with dislocations and disclinations. In sharp difference with other anomalies, there exist torsional anomalies depending on the position of Weyl nodes in the energy-momentum space. These anomalies originate from particles pumped up through the Weyl nodes and they are thus insensitive to the ultra-violet physics, while the Nieh-Yan anomaly is from the particle inflow through the ultra-violet cut-off. The current non-conservation as well as the energy-momentum non-conservation are found, which stem from the zero modes trapped in the dislocations and they can be understood from the Callan-Harvey mechanism. Finally, by comparing our results with the well-established momentum anomaly in the $^{3}\text{He-A}$ phase, the Nieh-Yan term as well as other cut-off dependent terms are shown to be negligible, because the ratio between the Lorentz symmetry breaking scale and the chemical potential is of order $10^{-5}$.

cond-mat.mes-hall↗

arXiv2020

Multi-Class Classification from Noisy-Similarity-Labeled Data

A similarity label indicates whether two instances belong to the same class while a class label shows the class of the instance. Without class labels, a multi-class classifier could be learned from similarity-labeled pairwise data by meta classification learning. However, since the similarity label is less informative than the class label, it is more likely to be noisy. Deep neural networks can easily remember noisy data, leading to overfitting in classification. In this paper, we propose a method for learning from only noisy-similarity-labeled data. Specifically, to model the noise, we employ a noise transition matrix to bridge the class-posterior probability between clean and noisy data. We further estimate the transition matrix from only noisy data and build a novel learning system to learn a classifier which can assign noise-free class labels for instances. Moreover, we theoretically justify how our proposed method generalizes for learning classifiers. Experimental results demonstrate the superiority of the proposed method over the state-of-the-art method on benchmark-simulated and real-world noisy-label datasets.

cs.LG↗

arXiv2020

Beyond Unfolding: Exact Recovery of Latent Convex Tensor Decomposition under Reshuffling

Exact recovery of tensor decomposition (TD) methods is a desirable property in both unsupervised learning and scientific data analysis. The numerical defects of TD methods, however, limit their practical applications on real-world data. As an alternative, convex tensor decomposition (CTD) was proposed to alleviate these problems, but its exact-recovery property is not properly addressed so far. To this end, we focus on latent convex tensor decomposition (LCTD), a practically widely-used CTD model, and rigorously prove a sufficient condition for its exact-recovery property. Furthermore, we show that such property can be also achieved by a more general model than LCTD. In the new model, we generalize the classic tensor (un-)folding into reshuffling operation, a more flexible mapping to relocate the entries of the matrix into a tensor. Armed with the reshuffling operations and exact-recovery property, we explore a totally novel application for (generalized) LCTD, i.e., image steganography. Experimental results on synthetic data validate our theory, and results on image steganography show that our method outperforms the state-of-the-art methods.

cs.LG↗

arXiv2019

Are Anchor Points Really Indispensable in Label-Noise Learning?

In label-noise learning, \textit{noise transition matrix}, denoting the probabilities that clean labels flip into noisy labels, plays a central role in building \textit{statistically consistent classifiers}. Existing theories have shown that the transition matrix can be learned by exploiting \textit{anchor points} (i.e., data points that belong to a specific class almost surely). However, when there are no anchor points, the transition matrix will be poorly learned, and those current consistent classifiers will significantly degenerate. In this paper, without employing anchor points, we propose a \textit{transition-revision} ($T$-Revision) method to effectively learn transition matrices, leading to better classifiers. Specifically, to learn a transition matrix, we first initialize it by exploiting data points that are similar to anchor points, having high \textit{noisy class posterior probabilities}. Then, we modify the initialized matrix by adding a \textit{slack variable}, which can be learned and validated together with the classifier by using noisy data. Empirical results on benchmark-simulated and real-world label-noise datasets demonstrate that without using exact anchor points, the proposed method is superior to the state-of-the-art label-noise learning methods.

cs.LG↗

arXiv2019

The general fifth-order nonlinear Schrödinger equation with nonzero boundary conditions: Inverse scattering transform and multisoliton solutions

Under investigation in this work is the inverse scattering transform of the general fifth-order nonlinear Schrödinger equation with nonzero boundary conditions (NZBCs), which can be reduced to several integrable equations. Firstly, a matrix Riemann-Hilbert problem for the equation with NZBCs at infinity is systematically investigated.Then the inverse problems are solved through the investigation of the matrix Riemann-Hilbert problem. Thus, the general solutions for the potentials, and explicit expressions for the reflection-less potentials are well constructed. Furthermore, the trace formulae and theta conditions are also presented. In particular, we analyze the simple-pole and double-pole solutions for the equation with NZBCs. Finally, the dynamics of the obtained solutions are graphically discussed. These results provided in this work can be useful to enrich and explain the related nonlinear wave phenomena in nonlinear fields.

math-ph↗

arXiv2019

Inverse scattering transform of an extended nonlinear Schrödinger equation with nonzero boundary conditions and its multisoliton solutions

Under investigation in this work is an extended nonlinear Schrödinger equation with nonzero boundary conditions, which can model the propagation of waves in dispersive media. Firstly, a matrix Riemann-Hilbert problem for the equation with nonzero boundary conditions at infinity is systematically discussed. Then the inverse problems are solved through the investigation of the matrix Riemann-Hilbert problem. Therefore, the general solutions for the potentials, and explicit expressions for the reflection-less potentials are presented. Furthermore, we construct the simple-pole and double-pole solutions for the equation. Finally, the remarkable characteristics of these solutions are graphically discussed. Our results should be useful to enrich and explain the related nonlinear wave phenomena in nonlinear fields.

math-ph↗

arXiv2019

Correlating the Electronic Structures of Metallic/Semiconductor MoTe2 Interface to its Atomic Structures

Contact interface properties are important in determining the performances of devices based on atomically thin two-dimensional (2D) materials, especially those with short channels. Understanding the contact interface is therefore quite important to design better devices. Herein, we use scanning transmission electron microscopy, electron energy loss spectroscopy, and first-principles calculations to reveal the electronic structures within the metallic (1T')-semiconducting (2H) MoTe2 coplanar phase boundary across a wide spectral range and correlate its properties and atomic structure. We find that the 2H-MoTe2 excitonic peaks cross the phase boundary into the 1T' phase within a range of approximately 150 nm. The 1T'-MoTe2 crystal field can penetrate the boundary and extend into the 2H phase by approximately two unit cells. The plasmonic oscillations exhibit strong angle dependence, i.e., a red-shift (approximately 0.3 eV-1.2 eV) occurs within 4 nm at 1T'/2H-MoTe2 boundaries with large tilt angles, but there is no shift at zero-tilted boundaries. These atomic-scale measurements reveal the structure-property relationships of 1T'/2H-MoTe2 boundary, providing useful information for phase boundary engineering and device development based on 2D materials.

cond-mat.mtrl-sci↗

arXiv2019

Where is the Bottleneck of Adversarial Learning with Unlabeled Data?

Deep neural networks (DNNs) are incredibly brittle due to adversarial examples. To robustify DNNs, adversarial training was proposed, which requires large-scale but well-labeled data. However, it is quite expensive to annotate large-scale data well. To compensate for this shortage, several seminal works are utilizing large-scale unlabeled data. In this paper, we observe that seminal works do not perform well, since the quality of pseudo labels on unlabeled data is quite poor, especially when the amount of unlabeled data is significantly larger than that of labeled data. We believe that the quality of pseudo labels is the bottleneck of adversarial learning with unlabeled data. To tackle this bottleneck, we leverage deep co-training, which trains two deep networks and encourages two networks diverged by exploiting peer's adversarial examples. Based on deep co-training, we propose robust co-training (RCT) for adversarial learning with unlabeled data. We conduct comprehensive experiments on CIFAR-10 and SVHN datasets. Empirical results demonstrate that our RCT can significantly outperform baselines (e.g., robust self-training (RST)) in both standard test accuracy and robust test accuracy w.r.t. different datasets, different network structures, and different types of adversarial training.

cs.LG↗

arXiv2019

A fast two-point gradient algorithm based on sequential subspace optimization method for nonlinear ill-posed problems

In this paper, we propose and analyze a fast two-point gradient algorithm for solving nonlinear ill-posed problems, which is based on the sequential subspace optimization method. A complete convergence analysis is provided under the classical assumptions for iterative regularization methods. The design of the two-point gradient method involves the choices of the combination parameters which is systematically discussed. Furthermore, detailed numerical simulations are presented for inverse potential problem, which exhibit that the proposed method leads to a strongly decrease of the iteration numbers and the overall computational time can be significantly reduced.

math.AP↗

arXiv2019

The nonlinear steepest descent approach for long time behavior of the two-component coupled Sasa-Satsuma equation with a $5\times 5$ Lax pair

Under investigation in this work is the coupled Sasa-Satsuma equation, which can describe the propagations of two optical pulse envelopes in birefringent fibers. The Riemann-Hilbert problem for the equation is formulated on the basis of the corresponding $5\times5$ matrix spectral problem, which allows us to present a suitable representation for the solution of the equation. Then the Deift-Zhou steepest descent method is used to analyze the long time behavior of the coupled Sasa-Satsuma equation.

math-ph↗

arXiv2019

SUM: Suboptimal Unitary Multi-task Learning Framework for Spatiotemporal Data Prediction

The typical multi-task learning methods for spatio-temporal data prediction involve low-rank tensor computation. However, such a method have relatively weak performance when the task number is small, and we cannot integrate it into non-linear models. In this paper, we propose a two-step suboptimal unitary method (SUM) to combine a meta-learning strategy into multi-task models. In the first step, it searches for a global pattern by optimising the general parameters with gradient descents under constraints, which is a geological regularizer to enable model learning with less training data. In the second step, we derive an optimised model on each specific task from the global pattern with only a few local training data. Compared with traditional multi-task learning methods, SUM shows advantages of generalisation ability on distant tasks. It can be applied on any multi-task models with the gradient descent as its optimiser regardless if the prediction function is linear or not. Moreover, we can harness the model to enable traditional prediction model to make coKriging. The experiments on public datasets have suggested that our framework, when combined with current multi-task models, has a conspicuously better prediction result when the task number is small compared to low-rank tensor learning, and our model has a quite satisfying outcome when adjusting the current prediction models for coKriging.

cs.LG↗

arXiv2019

VILD: Variational Imitation Learning with Diverse-quality Demonstrations

The goal of imitation learning (IL) is to learn a good policy from high-quality demonstrations. However, the quality of demonstrations in reality can be diverse, since it is easier and cheaper to collect demonstrations from a mix of experts and amateurs. IL in such situations can be challenging, especially when the level of demonstrators' expertise is unknown. We propose a new IL method called \underline{v}ariational \underline{i}mitation \underline{l}earning with \underline{d}iverse-quality demonstrations (VILD), where we explicitly model the level of demonstrators' expertise with a probabilistic graphical model and estimate it along with a reward function. We show that a naive approach to estimation is not suitable to large state and action spaces, and fix its issues by using a variational approach which can be easily implemented using existing reinforcement learning methods. Experiments on continuous-control benchmarks demonstrate that VILD outperforms state-of-the-art methods. Our work enables scalable and data-efficient IL under more realistic settings than before.

cs.LG↗

arXiv2019

A non-Abelian twist to integer quantum Hall states

Through a theoretical coupled wire model, we construct strongly correlated electronic \emph{integer} quantum Hall states. As a distinguishing feature, these states support electric and thermal Hall transport violating the Wiedemann-Franz law as $\left(κ_{xy}/σ_{xy}\right)/\left[\left(π^{2}k_{B}^{2}T\right)/3e^{2}\right]<1$.We propose a new Abelian incompressible fluid at filling $ν=16$ that supports a bosonic chiral $(E_{8})_{1}$ conformal field theory at the edge and is intimately related to topological paramagnets in (3+1)D. We further show that this topological phase can be partitioned into two non-Abelian quantum Hall states at filling $ν=8$, each carrying bosonic chiral $(G_{2})_{1}$ or $(F_{4})_{1}$ edge theories, and hosting Fibonacci anyonic excitations in the bulk. Finally, we discover a new notion of particle-hole conjugation based on the $E_{8}$ state that relates the $G_{2}$ and $F_{4}$ Fibonacci states.

cond-mat.str-el↗

arXiv2019

Towards Robust ResNet: A Small Step but A Giant Leap

This paper presents a simple yet principled approach to boosting the robustness of the residual network (ResNet) that is motivated by the dynamical system perspective. Namely, a deep neural network can be interpreted using a partial differential equation, which naturally inspires us to characterize ResNet by an explicit Euler method. Our analytical studies reveal that the step factor h in the Euler method is able to control the robustness of ResNet in both its training and generalization. Specifically, we prove that a small step factor h can benefit the training robustness for back-propagation; from the view of forward-propagation, a small h can aid in the robustness of the model generalization. A comprehensive empirical evaluation on both vision CIFAR-10 and text AG-NEWS datasets confirms that a small h aids both the training and generalization robustness.

cs.CV↗

arXiv2019

A coupled wire description of surface $ADE$ topological orders

Symmetry-protected and symmetry-enriched topological (SPT/SET) phases in three dimensions are quantum systems that support non-trivial two-dimensional surface states. These surface states develop finite excitation energy gaps when the relevant symmetries are broken. On the other hand, one-dimensional gapless modes can populate along interfaces that separate adjacent gapped surface domains with distinct symmetry-breaking orders. A surface strip pattern in general reduces the low-energy SPT/SET surface degrees of freedom onto a 2D array of gapless 1D channels. These channels can be coupled to one another by quasiparticle tunneling, and these inter-wire interactions collectively provide an effective description of the surface state. In this paper, we study a general class of symmetry-preserving or breaking SPT/SET surface states that admit finite excitation energy gaps and Abelian topological orders via the coupled wire construction. In particular, we focus on the prototype Abelian surface topological orders that fall under the $ADE$ classification of simply-laced Lie algebras. We also elaborate on the emergent symmetry and duality properties of the coupled wire models.

cond-mat.str-el↗

arXiv2019

How does Disagreement Help Generalization against Label Corruption?

Learning with noisy labels is one of the hottest problems in weakly-supervised learning. Based on memorization effects of deep neural networks, training on small-loss instances becomes very promising for handling noisy labels. This fosters the state-of-the-art approach "Co-teaching" that cross-trains two deep neural networks using the small-loss trick. However, with the increase of epochs, two networks converge to a consensus and Co-teaching reduces to the self-training MentorNet. To tackle this issue, we propose a robust learning paradigm called Co-teaching+, which bridges the "Update by Disagreement" strategy with the original Co-teaching. First, two networks feed forward and predict all data, but keep prediction disagreement data only. Then, among such disagreement data, each network selects its small-loss data, but back propagates the small-loss data from its peer network and updates its own parameters. Empirical results on benchmark datasets demonstrate that Co-teaching+ is much superior to many state-of-the-art methods in the robustness of trained models.

cs.LG↗