arXiv subjects

Yichen Xu

Publications and source records attributed to Yichen Xu.

68 records · Page 4Linked to original sources

arXiv2021

Structure-Aware Hard Negative Mining for Heterogeneous Graph Contrastive Learning

Recently, heterogeneous Graph Neural Networks (GNNs) have become a de facto model for analyzing HGs, while most of them rely on a relative large number of labeled data. In this work, we investigate Contrastive Learning (CL), a key component in self-supervised approaches, on HGs to alleviate the label scarcity problem. We first generate multiple semantic views according to metapaths and network schemas. Then, by pushing node embeddings corresponding to different semantic views close to each other (positives) and pulling other embeddings apart (negatives), one can obtain informative representations without human annotations. However, this CL approach ignores the relative hardness of negative samples, which may lead to suboptimal performance. Considering the complex graph structure and the smoothing nature of GNNs, we propose a structure-aware hard negative mining scheme that measures hardness by structural characteristics for HGs. By synthesizing more negative nodes, we give larger weights to harder negatives with limited computational overhead to further boost the performance. Empirical studies on three real-world datasets show the effectiveness of our proposed method. The proposed method consistently outperforms existing state-of-the-art methods and notably, even surpasses several supervised counterparts.

cs.LG↗

arXiv2021

Disentangled Self-Attentive Neural Networks for Click-Through Rate Prediction

Click-Through Rate (CTR) prediction, whose aim is to predict the probability of whether a user will click on an item, is an essential task for many online applications. Due to the nature of data sparsity and high dimensionality of CTR prediction, a key to making effective prediction is to model high-order feature interaction. An efficient way to do this is to perform inner product of feature embeddings with self-attentive neural networks. To better model complex feature interaction, in this paper we propose a novel DisentanglEd Self-atTentIve NEtwork (DESTINE) framework for CTR prediction that explicitly decouples the computation of unary feature importance from pairwise interaction. Specifically, the unary term models the general importance of one feature on all other features, whereas the pairwise interaction term contributes to learning the pure impact for each feature pair. We conduct extensive experiments using two real-world benchmark datasets. The results show that DESTINE not only maintains computational efficiency but achieves consistent improvements over state-of-the-art baselines.

cs.IR↗

arXiv2021

Green's function Zero and Symmetric Mass Generation

It is known that, under short-range interactions many topological superconductors (TSC) and topological insulators (TI) are trivialized, which means the boundary state of the system can be trivially gapped out by interaction without leading to symmetry breaking or topological ground state degeneracy. This phenomenon is also referred to as "symmetric mass generation" (SMG), and has attracted broad attentions from both the condensed matter and high energy physics communities. However, after the trivialization caused by interaction, some trace of the nontrivial topology of the system still persists. Previous studies have indicated that interacting topological TSC and TI could be related to the "zero" of Green's function, namely the fermion Green's function $G(\mathrm{i} ω\rightarrow 0) = 0$. In this work, through the general "decorated defect" construction of symmetry protected topological (SPT) states, we demonstrate the existence of Green's function zero after SMG, by mapping the evaluation of the Green's function to the problem of a single particle path integral. This method can be extended to the cases without spatial translation symmetry, where the momentum space which hosts many quantized topological numbers is no longer meaningful. Using the same method one can demonstrate the existence of the Green's function zero at the "avoided topological transition" in the bulk of the system.

cond-mat.str-el↗

arXiv2021

Graph Contrastive Learning with Adaptive Augmentation

Recently, contrastive learning (CL) has emerged as a successful method for unsupervised graph representation learning. Most graph CL methods first perform stochastic augmentation on the input graph to obtain two graph views and maximize the agreement of representations in the two views. Despite the prosperous development of graph CL methods, the design of graph augmentation schemes -- a crucial component in CL -- remains rarely explored. We argue that the data augmentation schemes should preserve intrinsic structures and attributes of graphs, which will force the model to learn representations that are insensitive to perturbation on unimportant nodes and edges. However, most existing methods adopt uniform data augmentation schemes, like uniformly dropping edges and uniformly shuffling features, leading to suboptimal performance. In this paper, we propose a novel graph contrastive representation learning method with adaptive augmentation that incorporates various priors for topological and semantic aspects of the graph. Specifically, on the topology level, we design augmentation schemes based on node centrality measures to highlight important connective structures. On the node attribute level, we corrupt node features by adding more noise to unimportant node features, to enforce the model to recognize underlying semantic information. We perform extensive experiments of node classification on a variety of real-world datasets. Experimental results demonstrate that our proposed method consistently outperforms existing state-of-the-art baselines and even surpasses some supervised counterparts, which validates the effectiveness of the proposed contrastive framework with adaptive augmentation.

cs.LG↗

arXiv2020

Continuous Néel-VBS Quantum Phase Transition in Non-Local one-dimensional systems with SO(3) Symmetry

One dimensional $(1d)$ interacting systems with local Hamiltonians can be studied with various well-developed analytical methods. Recently novel $1d$ physics was found numerically in systems with either spatially nonlocal interactions, or at the $1d$ boundary of $2d$ quantum critical points, and the critical fluctuation in the bulk also yields effective nonlocal interactions at the boundary. This work studies the edge states at the $1d$ boundary of $2d$ strongly interacting symmetry protected topological (SPT) states, when the bulk is driven to a disorder-order phase transition. We will take the $2d$ Affleck-Kennedy-Lieb-Tasaki (AKLT) state as an example, which is a SPT state protected by the SO(3) spin symmetry and spatial translation. We found that the original $(1+1)d$ boundary conformal field theory of the AKLT state is unstable due to coupling to the boundary avatar of the bulk quantum critical fluctuations. When the bulk is fixed at the quantum critical point, within the accuracy of our expansion method, we find that by tuning one parameter at the boundary, there is a generic direct transition between the long range antiferromagnetic Néel order and the valence bond solid (VBS) order. This transition is very similar to the Néel-VBS transition recently found in numerical simulation of a spin-1/2 chain with nonlocal spatial interactions. Connections between our analytical studies and recent numerical results concerning the edge states of the $2d$ AKLT-like state at a bulk quantum phase transition will also be discussed.

cond-mat.str-el↗

arXiv2020

CAGNN: Cluster-Aware Graph Neural Networks for Unsupervised Graph Representation Learning

Unsupervised graph representation learning aims to learn low-dimensional node embeddings without supervision while preserving graph topological structures and node attributive features. Previous graph neural networks (GNN) require a large number of labeled nodes, which may not be accessible in real-world graph data. In this paper, we present a novel cluster-aware graph neural network (CAGNN) model for unsupervised graph representation learning using self-supervised techniques. In CAGNN, we perform clustering on the node embeddings and update the model parameters by predicting the cluster assignments. Moreover, we observe that graphs often contain inter-class edges, which mislead the GNN model to aggregate noisy information from neighborhood nodes. We further refine the graph topology by strengthening intra-class edges and reducing node connections between different classes based on cluster labels, which better preserves cluster structures in the embedding space. We conduct comprehensive experiments on two benchmark tasks using real-world datasets. The results demonstrate the superior performance of the proposed model over existing baseline methods. Notably, our model gains over 7% improvements in terms of accuracy on node clustering over state-of-the-arts.

cs.LG↗

arXiv2020

Physics of Symmetry Protected Topological phases involving Higher Symmetries and their Applications

We discuss physical constructions, and the boundary properties of various symmetry protected topological phases that involve 1-form symmetries, from one spatial dimension (1d) to four spatial dimensions (4d). For example, the prototype 3d boundary state of 4d SPT states involving 1-form symmetries can be either a gapless photon phase (quantum electrodynamics) or gapped topological order enriched by 1-form symmetries, namely the loop excitations of these topological orders carry nontrivial 1-form symmetry charges. This study also serves the purpose of diagnosing anomaly of 3d states of matter. Connection between SPT states with 1-form symmetries and condensed matter systems such as quantum dimer models at one lower dimension will also be discussed. Whether a quantum dimer model can have a trivial gapped phase or not depends on the nature of its corresponding bulk state in one higher dimension.

cond-mat.str-el↗

arXiv2020

Deep Graph Contrastive Representation Learning

Graph representation learning nowadays becomes fundamental in analyzing graph-structured data. Inspired by recent success of contrastive methods, in this paper, we propose a novel framework for unsupervised graph representation learning by leveraging a contrastive objective at the node level. Specifically, we generate two graph views by corruption and learn node representations by maximizing the agreement of node representations in these two views. To provide diverse node contexts for the contrastive objective, we propose a hybrid scheme for generating graph views on both structure and attribute levels. Besides, we provide theoretical justification behind our motivation from two perspectives, mutual information and the classical triplet loss. We perform empirical experiments on both transductive and inductive learning tasks using a variety of real-world datasets. Experimental experiments demonstrate that despite its simplicity, our proposed method consistently outperforms existing state-of-the-art methods by large margins. Moreover, our unsupervised method even surpasses its supervised counterparts on transductive tasks, demonstrating its great potential in real-world applications.

cs.LG↗

arXiv2020

Orbital Orders and non-Fermi Liquid in Moiré systems

Motivated by recent observation of nematicity in Moiré systems, we study three different orbital orders that potentially can happen in Moiré systems: (1) the nematic order; (2) the valley polarization; and (3) the "compass order". Each order parameter spontaneously breaks part of the spatial symmetries of the system. We explore physics caused by the quantum fluctuations close to the order-disorder transition of these order parameters. Especially, we recognize that the symmetry of the Moiré systems leads to a crucial difference of the effective theory describing the nematic order from the standard Hertz-Millis formalism. We demonstrate that this key difference may lead to a special non-Fermi liquid behavior near the order-disorder nematic transition, different from the standard non-Fermi liquid behavior usually expected when a Fermi surface is coupled to the critical fluctuations of orbital orders. We also discuss the interplay of the three order parameters and the possible rich phase diagram at finite temperature. Within the three orbital orders, the valley polarization and the "compass order" likely strongly compete with the superconductor.

cond-mat.str-el↗

arXiv2020

Non-Landau Quantum Phase Transitions and nearly-Marginal non-Fermi Liquid

Non-fermi liquid and unconventional quantum critical points (QCP) with strong fractionalization are two exceptional phenomena beyond the classic condensed matter doctrines, both of which could occur in strongly interacting quantum many-body systems. This work demonstrates that using a controlled method one can construct a non-Fermi liquid within a considerable energy window based on the unique physics of unconventional QCPs. We will focus on the "nearly-marginal non-Fermi liquid", defined as a state whose fermion self-energy scales as $Σ_f(i ω) \sim i \mathrm{sgn}(ω)|ω|^α$ with $α$ close to $1$ in a considerable energy window. The nearly-marginal non-fermi liquid is obtained by coupling an electron fermi surface to unconventional QCPs that are beyond the Landau's paradigm. This mechanism relies on the observation that the anomalous dimension $η$ of the order parameter of these unconventional QCPs can be close to $1$, which is significantly larger than conventional Landau phase transitions, for example the Wilson-Fisher fixed points. The fact that $η\sim 1$ justifies a perturbative renormalization group calculation proposed earlier. Various candidate QCPs that meet this desired condition are proposed.

cond-mat.str-el↗

arXiv2020

Topological Edge and Interface states at Bulk disorder-to-order Quantum Critical Points

We study the interplay between two nontrivial boundary effects: (1) the two dimensional ($2d$) edge states of three dimensional ($3d$) strongly interacting bosonic symmetry protected topological states, and (2) the boundary fluctuations of $3d$ bulk disorder-to-order phase transitions. We then generalize our study to $2d$ gapless states localized at an interface embedded in a $3d$ bulk, when the bulk undergoes a quantum phase transition. Our study is based on generic long wavelength descriptions of these systems and controlled analytic calculations. Our results are summarized as follows: ($i.$) The edge state of a prototype bosonic symmetry protected states can be driven to a new fixed point by coupling to the boundary fluctuations of a bulk quantum phase transition; ($ii.$) the states localized at a $2d$ interface of a $3d$ SU(N) quantum antiferromagnet may be driven to a new fixed point by coupling to the bulk quantum critical modes. Properties of the new fixed points identified are also studied.

cond-mat.str-el↗

arXiv2020

Boundary Criticality of Topological Quantum Phase Transitions in $2d$ systems

We discuss the boundary critical behaviors of two dimensional quantum phase transitions with fractionalized degrees of freedom in the bulk, motivated by the fact that usually it is the $1d$ boundary that is exposed and can be conveniently probed in many experimental platforms. In particular, we mainly discuss boundary criticality of two examples: i. the quantum phase transition between a $2d$ $Z_2$ topological order and an ordered phase with spontaneous symmetry breaking; ii. the continuous quantum phase transition between metal and a particular type of Mott insulator (U(1) spin liquid). This theoretical study could be relevant to many purely $2d$ systems, where recent experiments have found correlated insulator, superconductor, and metal in the same phase diagram.

cond-mat.str-el↗

arXiv2019

Interacting Valley Chern Insulator and its Topological Imprint on Moiré Superconductors

One salient feature of systems with Moiré superlattice is that, Chern number of "minibands" originating from each valley of the original graphene Brillouin zone becomes a well-defined quantized number because the miniband from each valley can be isolated from the rest of the spectrum due to the Moiré potential. Then a Moiré system with a well-defined valley Chern number can become a nonchiral topological insulator with $U(1) \times Z_3 $ symmetry and a $\mathbb{Z}$ classification at the free fermion level. Here we demonstrate that the strongly interacting nature of the Moiré system reduces the classification of the valley Chern insulator from $\mathbb{Z}$ to $\mathbb{Z}_3$, and it is topologically equivalent to a bosonic symmetry protected topological state made of local boson operators. We also demonstrate that, even if the system becomes a superconductor when doped away from the valley Chern insulator, the valley Chern insulator still leaves a topological imprint as the localized Majorana fermion zero mode in certain geometric configuration.

cond-mat.str-el↗

arXiv2017

Topological Corner States on Kagome Lattice Based Chiral Higher-Order Topological Insulator

The higher-order topological insulator (HOTI) protected by spacial symmetry has been studied in-depth on models with square lattice. Our work, based on an alternative model on the breathing Kagome lattice, revealed that the different types of corners in the lattice could actually be conditionally gapless, or always gapped. Using the Wilson loop formalism, we argue that these corner states occur when the eigenvalues of the Wannier Hamiltonian cross through a certain reference point during the conceptual "pumping" procedure. The results demonstrate the corner of the Kagome lattice based HOTI is a zero-dimensional analogue of the 1D chiral edge states on the boundary of a Chern insulator, but with a sensitive dependence on the shape of the corner. Our method of the pumping cylinder, which reveals the symmetry/gapless-ability correspondence, can be generalized into a general scheme in determining the classification of corner(hinge) states in HOTI.

cond-mat.str-el↗