arXiv subjects

Zhen Bi

Publications and source records attributed to Zhen Bi.

At least 73 records · Page 4Linked to original sources

arXiv2021

Improving Knowledge Graph Representation Learning by Structure Contextual Pre-training

Representation learning models for Knowledge Graphs (KG) have proven to be effective in encoding structural information and performing reasoning over KGs. In this paper, we propose a novel pre-training-then-fine-tuning framework for knowledge graph representation learning, in which a KG model is firstly pre-trained with triple classification task, followed by discriminative fine-tuning on specific downstream tasks such as entity type prediction and entity alignment. Drawing on the general ideas of learning deep contextualized word representations in typical pre-trained language models, we propose SCoP to learn pre-trained KG representations with structural and contextual triples of the target triple encoded. Experimental results demonstrate that fine-tuning SCoP not only outperforms results of baselines on a portfolio of downstream tasks but also avoids tedious task-specific model design and parameter training.

cs.AI↗

arXiv2021

Normal vs. Adversarial: Salience-based Analysis of Adversarial Samples for Relation Extraction

Recent neural-based relation extraction approaches, though achieving promising improvement on benchmark datasets, have reported their vulnerability towards adversarial attacks. Thus far, efforts mostly focused on generating adversarial samples or defending adversarial attacks, but little is known about the difference between normal and adversarial samples. In this work, we take the first step to leverage the salience-based method to analyze those adversarial samples. We observe that salience tokens have a direct correlation with adversarial perturbations. We further find the adversarial perturbations are either those tokens not existing in the training set or superficial cues associated with relation labels. To some extent, our approach unveils the characters against adversarial samples. We release an open-source testbed, "DiagnoseAdv" in https://github.com/zjunlp/DiagnoseAdv.

cs.CL↗

arXiv2021

Yang-Lee edge singularity triggered entanglement transition

We show that a class of $\mathcal{PT}$ symmetric non-Hermitian Hamiltonians realizing the Yang-Lee edge singularity exhibits an entanglement transition in the long-time steady state evolved under the Hamiltonian. Such a transition is induced by a level crossing triggered by the critical point associated with the Yang-Lee singularity and hence is first-order in nature. At the transition, the entanglement entropy of the steady state jumps discontinuously from a volume-law to an area-law scaling. We exemplify this mechanism using a one-dimensional transverse field Ising model with additional imaginary fields, as well as the spin-1 Blume-Capel model and the three-state Potts model. We further make a connection to the forced-measurement induced entanglement transition in a Floquet non-unitary circuit subject to continuous measurements followed by post-selections. Our results demonstrate a new mechanism for entanglement transitions in non-Hermitian systems harboring a critical point.

cond-mat.str-el↗

arXiv2021

Learning to Ask for Data-Efficient Event Argument Extraction

Event argument extraction (EAE) is an important task for information extraction to discover specific argument roles. In this study, we cast EAE as a question-based cloze task and empirically analyze fixed discrete token template performance. As generating human-annotated question templates is often time-consuming and labor-intensive, we further propose a novel approach called "Learning to Ask," which can learn optimized question templates for EAE without human annotations. Experiments using the ACE-2005 dataset demonstrate that our method based on optimized questions achieves state-of-the-art performance in both the few-shot and supervised settings.

cs.CL↗

arXiv2021

Text-guided Legal Knowledge Graph Reasoning

Recent years have witnessed the prosperity of legal artificial intelligence with the development of technologies. In this paper, we propose a novel legal application of legal provision prediction (LPP), which aims to predict the related legal provisions of affairs. We formulate this task as a challenging knowledge graph completion problem, which requires not only text understanding but also graph reasoning. To this end, we propose a novel text-guided graph reasoning approach. We collect amounts of real-world legal provision data from the Guangdong government service website and construct a legal dataset called LegalLPP. Extensive experimental results on the dataset show that our approach achieves better performance compared with baselines. The code and dataset are available in \url{https://github.com/zxlzr/LegalPP} for reproducibility.

cs.AI↗

arXiv2021

Disentangled Contrastive Learning for Learning Robust Textual Representations

Although the self-supervised pre-training of transformer models has resulted in the revolutionizing of natural language processing (NLP) applications and the achievement of state-of-the-art results with regard to various benchmarks, this process is still vulnerable to small and imperceptible permutations originating from legitimate inputs. Intuitively, the representations should be similar in the feature space with subtle input permutations, while large variations occur with different meanings. This motivates us to investigate the learning of robust textual representation in a contrastive manner. However, it is non-trivial to obtain opposing semantic instances for textual samples. In this study, we propose a disentangled contrastive learning method that separately optimizes the uniformity and alignment of representations without negative sampling. Specifically, we introduce the concept of momentum representation consistency to align features and leverage power normalization while conforming the uniformity. Our experimental results for the NLP benchmarks demonstrate that our approach can obtain better results compared with the baselines, as well as achieve promising improvements with invariance tests and adversarial attacks. The code is available in https://github.com/zxlzr/DCL.

cs.CL↗

arXiv2021

Interventional Aspect-Based Sentiment Analysis

Recent neural-based aspect-based sentiment analysis approaches, though achieving promising improvement on benchmark datasets, have reported suffering from poor robustness when encountering confounder such as non-target aspects. In this paper, we take a causal view to addressing this issue. We propose a simple yet effective method, namely, Sentiment Adjustment (SENTA), by applying a backdoor adjustment to disentangle those confounding factors. Experimental results on the Aspect Robustness Test Set (ARTS) dataset demonstrate that our approach improves the performance while maintaining accuracy in the original test set.

cs.CL↗

arXiv2020

Excitonic density wave and spin-valley superfluid in bilayer transition metal dichalcogenide

Artificial moiré superlattices in 2d van der Waals heterostructure is a new venue for realizing and controlling correlated electronic phenomena. Recently, twisted bilayer WSe$_2$ emerged as a new robust moiré system hosting a correlated insulator at moiré half-filling over a range of twist angle. In this work, we present a theory of this insulating state as an excitonic density wave due to intervalley electron-hole pairing. We show that exciton condensation is strongly enhanced by a van Hove singularity near the Fermi level. Our theory explains the remarkable sensitivity of the insulating gap to the vertical electric field. In contrast, the gap is weakly reduced by a perpendicular magnetic field, with quadratic dependence at low field. The different responses to electric and magnetic field can be understood in terms of pair-breaking versus non-pair-breaking effects in a BCS analog of the system. We further predict superfluid spin transport in this electrical insulator, which can be detected by optical spin injection and spatial-temporal imaging.

cond-mat.str-el↗

arXiv2019

Lattice Analysis of $SU(2)$ with 1 Adjoint Dirac Flavor

Recently $SU(2)$ Yang-Mills theory with one massless adjoint Dirac quark flavor emerges as a novel critical theory that can describe the evolution between a trivial insulator and a topological insulator in AIII class in $3+1$ dimensions. There are several classes of conjectured infrared dynamics for this theory. One possibility is that the theory undergoes spontaneous chiral symmetry breaking, with two massless Goldstone bosons (the scalar diquark and its antiparticle) in the infrared. Another scenario, which is suggested by previous lattice studies by Athenodorou et al., is that the IR sector of the theory is a strongly interacting conformal field theory as the quark mass vanishes. The most recent theoretical proposals argue for a case that in the infrared a composite fermion composed of two quarks and an antiquark becomes massless and non-interacting as the quark mass goes to zero, while other sectors are decoupled from this low-energy fermion. This work expands upon previous studies by including the composite fermion to investigate which of these three potential scenarios captures the infrared behavior of this theory.

hep-lat↗

arXiv2019

Landau ordering phase transitions beyond the Landau paradigm

Continuous phase transitions associated with the onset of a spontaneously broken symmetry are thought to be successfully described by the Landau-Ginzburg-Wilson-Fisher theory of fluctuating order parameters. In this work we show that such transitions can admit new universality classes which cannot be understood in terms of a theory of order parameter fluctuations. We explicitly demonstrate continuous time reversal symmetry breaking quantum phase transitions of $3+1$-D bosonic systems described by critical theories expressed in terms of a deconfined gauge theory with massless Dirac fermions instead of the fluctuating Ising order parameter. We dub such phase transitions "Landau transitions beyond Landau description" (LBL). A key feature of our examples is that the stability of the LBL fixed points requires a crucial global symmetry, which is non-anomalous, unbroken, and renders no symmetry protected topological phase throughout the phase diagram. Despite this, there are elementary critical fluctuations of the phase transition that transform projectively under this symmetry group. We also construct examples of other novel quantum critical phenomena, notably a continuous Landau-forbidden deconfined critical point between two Landau-allowed phases in $3+1$-D.

cond-mat.str-el↗

arXiv2019

Fractonic Matter in Symmetry-Enriched U(1) Gauge Theory

In this work we explore the interplay between global symmetry and the mobility of quasiparticle excitations. We show that fractonic matter naturally appears in a three dimensional U(1) gauge theory, enriched by global U(1) and translational symmetries, via the mechanism of anyonic spin-orbital coupling. We develop a systematic understanding of such symmetry-enforced mobility restrictions in terms of the classification of U(1) gauge theories enriched by U(1) and translational symmetries. We provide a unified construction of these phases by gauging layered symmetry-protected topological phases.

cond-mat.str-el↗

arXiv2019

Collusion of Interactions and Disorder at the Superfluid-Insulator Transition: A Dirty 2d Quantum Critical Point

We study the stability of the Wilson-Fisher fixed point of the quantum $\mathrm{O}(2N)$ vector model to quenched disorder in the large-$N$ limit. While a random mass is strongly relevant at the Gaussian fixed point, its effect is screened by the strong interactions of the Wilson-Fisher fixed point. This enables a perturbative renormalization group study of the interplay of disorder and interactions about this fixed point. We show that, in contrast to the spiralling flows obtained in earlier double-$ε$ expansions, the theory flows directly to a quantum critical point characterized by finite disorder and interactions. The critical exponents we obtain for this transition are in remarkable agreement with numerical studies of the superfluid-Mott glass transition. We additionally discuss the stability of this fixed point to scalar and vector potential disorder and use proposed boson-fermion dualities to make conjectures regarding the effects of weak disorder on dual Abelian Higgs and Chern-Simons-Dirac fermion theories when $N=1$.

cond-mat.str-el↗

arXiv2019

Superconductivity near a ferroelectric quantum critical point in ultralow-density Dirac materials

The experimental observation of superconductivity in doped semimetals and semiconductors, where the Fermi energy is comparable to or smaller than the characteristic phonon frequencies, is not captured by the conventional theory. In this paper, we propose a mechanism for superconductivity in ultralow-density three-dimensional Dirac materials based on the proximity to a ferroelectric quantum critical point. We derive a low-energy theory that takes into account both the strong Coulomb interaction and the direct coupling between the electrons and the soft phonon modes. We show that the Coulomb repulsion is strongly screened by the lattice polarization near the critical point even in the case of vanishing carrier density. Using a renormalization group analysis, we demonstrate that the effective electron-electron interaction is dominantly mediated by the transverse phonon mode. We find that the system generically flows towards strong electron-phonon coupling. Hence, we propose a new mechanism to simultaneously produce an attractive interaction and suppress strong Coulomb repulsion, which does not require retardation. For comparison, we perform same analysis for covalent crystals, where lattice polarization is negligible. We obtain qualitatively similar results, though the screening of the Coulomb repulsion is much weaker. We then apply our results to study superconductivity in the low-density limit. We find strong enhancement of the transition temperature upon approaching the quantum critical point. Finally, we also discuss scenarios to realize a topological $p$-wave superconducting state in covalent crystals close to the critical point.

cond-mat.supr-con↗

arXiv2019

Emergent fractons and algebraic quantum liquid from plaquette melting transitions

Paramagnetic spin systems with spontaneously broken spatial symmetries, such as valence bond solid (VBS) phases, can host topological defects carrying non-trivial quantum numbers, which enables the paradigm of deconfined quantum criticality. In this work, we study the properties of topological defects in valence plaquette solid (VPS) phases on square and cubic lattices. We show that the defects of the VPS order parameter, in addition to possessing non-trivial quantum numbers, have fracton mobility constraints deep in the VPS phase, which has been overlooked previously. The spinon inside a single vortex cannot move freely in any direction, while a dipolar pair of vortices with spinon pairs can only move perpendicular to its dipole moment. These mobility constraints, while they persist, can potentially inhibit the condensation of vortices and preclude a continuous transition from the VPS to the Néel antiferromagnet. Instead, the VPS melting transition can be driven by proliferation of spinon dipoles. For example, we argue that a $2d$ VPS can melt into a stable gapless phase in the form of an algebraic bond liquid with algebraic correlations and long range entanglement. Such a bond liquid phase yields a concrete example of the elusive $2d$ Bose metal with symmetry fractionalization. We also study $3d$ valence plaquette and valence cube ordered phase, and demonstrate that the topological defects therein also have fractonic dynamics. Possible nearby phases after melting the valence plaquettes or cubes are also discussed.

cond-mat.str-el↗

arXiv2019

Designing Flat Band by Strain

We study the effects of heterostrain on moiré bands in twisted bilayer graphene and bilayer transition metal dichalcogenide (TMD) systems. For bilayer graphene with twist angle near $1^\circ$, we show that heterostrain significantly increases the energy separation between conduction and valence bands as well as the Dirac velocity at charge neutrality, which resolves several puzzles in scanning tunneling spectroscopy and quantum oscillation experiments at once. For bilayer TMD, we show that applying small heterostrain generally leads to flat moiré bands that are highly tunable.

cond-mat.mtrl-sci↗

arXiv2018

An Adventure in Topological Phase Transitions in 3 + 1-D: Non-abelian Deconfined Quantum Criticalities and a Possible Duality

Continuous quantum phase transitions that are beyond the conventional paradigm of fluctuations of a symmetry breaking order parameter are challenging for theory. These phase transitions often involve emergent deconfined gauge fields at the critical points as demonstrated in 2+1-dimensions. Examples include phase transitions in quantum magnetism as well as those between Symmetry Protected Topological phases. In this paper, we present several examples of Deconfined Quantum Critical Points (DQCP) between Symmetry Protected Topological phases in 3+1-D for both bosonic and fermionic systems. Some of the critical theories can be formulated as non-abelian gauge theories either in their Infra-Red free regime, or in the conformal window when they flow to the Banks-Zaks fixed points. We explicitly demonstrate several interesting quantum critical phenomena. We describe situations in which the same phase transition allows for multiple universality classes controlled by distinct fixed points. We exhibit the possibility - which we dub "unnecessary quantum critical points" - of stable generic continuous phase transitions within the same phase. We present examples of interaction driven band-theory- forbidden continuous phase transitions between two distinct band insulators. The understanding we develop leads us to suggest an interesting possible 3+1-D field theory duality between SU(2) gauge theory coupled to one massless adjoint Dirac fermion and the theory of a single massless Dirac fermion augmented by a decoupled topological field theory.

cond-mat.str-el↗

arXiv2018

Lattice Construction of Duality with Non-Abelian Gauge Fields in 2+1D

The lattice construction of Euclidean path integrals has been a successful approach of deriving 2+1D field theory dualities with a U$(1)$ gauge field. In this work, we generalize this lattice construction to dualities with non-Abelian gauge fields. We construct the Euclidean spacetime lattice path integral for a theory with strongly-interacting $SO(3)$ vector bosons and Majorana fermions coupled to an $SO(3)$ gauge field and derive an exact duality between this theory and the theory of a free Majorana fermion on the spacetime lattice. We argue that this lattice duality implies the desired infrared duality between the field theory with an $SO(3)$ vector critical boson coupled to an $SO(3)_1$ Chern-Simons gauge theory, and a free massless Majorana fermion in 2+1D. We also generalize the lattice construction of dualities to models with $O(3)$ gauge fields.

cond-mat.str-el↗

arXiv2018

Classification of Symmetry-Protected Phases for Interacting Fermions in Two Dimensions

Recently, it has been shown that two-dimensional bosonic symmetry-protected topological(SPT) phases with on-site unitary symmetry $G$ can be completely classified by the group cohomology class $H^3(G, \mathrm{U}(1))$. Later, group super-cohomology class was proposed as a partial classification for SPT phases of interacting fermions. In this work, we revisit this problem based on the mathematical framework of $G$-extension of unitary braided tensor category(UBTC) theory. We first reproduce the partial classifications given by group super-cohomology, then we show that with an additional $H^1(G, \mathbb{Z}_2)$ structure, a complete classification of SPT phases for two-dimensional interacting fermion systems for a total symmetry group $G\times\mathbb{Z}_2^f$ can be achieved. We also discuss the classification of interacting fermionic SPT phases protected by time-reversal symmetry.

cond-mat.str-el↗