SearcharxivSearch

arXiv subjects

Zheng Hua

Publications and source records attributed to Zheng Hua.

At least 19 recordsLinked to original sources

Feigin-Odesskii brackets associated with Kodaira cycles and positroid varieties

We establish a link between open positroid varieties in the Grassmannians $G(k,n)$ and certain moduli spaces of complexes of vector bundles over Kodaira cycle $C^n$, using the shifted Poisson structure on the latter moduli spaces and relating them to a certain twist of the standard Poisson structure on $G(k,n)$. %by a bivector field on its maximal torus. This link allows us to solve a classification problem for extensions of vector bundles over $C^n$. Based on this solution we further classify the symplectic leaves of all positroid varieties in $G(k,n)$ with respect to the twisted standard Poisson structure. Moreover, we get an explicit description of the moduli stack of symplectic leaves of $G(k,n)$ with the twisted standard Poisson structure as an open substack of the stack of vector bundles on $C^n$.

math.AG

Exactness and Effective Degree Bound of Lasserre's Relaxation for Polynomial Optimization over Finite Variety

In this paper, we address the effective degree bound problem for Lasserre's hierarchy of moment-sum-of-squares (SOS) relaxations in polynomial optimization involving $n$ variables. We assume that the first $n$ equality constraint polynomials $g_1,\ldots,g_n$ do not share any nontrivial common complex zero locus at infinity and that the optimal solutions are nonsingular. Under these conditions, we derive an effective degree bound for the exactness of Lasserre's hierarchy. Importantly, the assumption of no solutions at infinity holds on a Zariski open set within the space of polynomials of fixed degrees. As a direct consequence, we provide the first explicit degree bound for gradient-type SOS relaxation under a generic condition.

math.OC

CL-CaGAN: Capsule differential adversarial continuous learning for cross-domain hyperspectral anomaly detection

Anomaly detection (AD) has attracted remarkable attention in hyperspectral image (HSI) processing fields, and most existing deep learning (DL)-based algorithms indicate dramatic potential for detecting anomaly samples through specific training process under current scenario. However, the limited prior information and the catastrophic forgetting problem indicate crucial challenges for existing DL structure in open scenarios cross-domain detection. In order to improve the detection performance, a novel continual learning-based capsule differential generative adversarial network (CL-CaGAN) is proposed to elevate the cross-scenario learning performance for facilitating the real application of DL-based structure in hyperspectral AD (HAD) task. First, a modified capsule structure with adversarial learning network is constructed to estimate the background distribution for surmounting the deficiency of prior information. To mitigate the catastrophic forgetting phenomenon, clustering-based sample replay strategy and a designed extra self-distillation regularization are integrated for merging the history and future knowledge in continual AD task, while the discriminative learning ability from previous detection scenario to current scenario is retained by the elaborately designed structure with continual learning (CL) strategy. In addition, the differentiable enhancement is enforced to augment the generation performance of the training data. This further stabilizes the training process with better convergence and efficiently consolidates the reconstruction ability of background samples. To verify the effectiveness of our proposed CL-CaGAN, we conduct experiments on several real HSIs, and the results indicate that the proposed CL-CaGAN demonstrates higher detection performance and continuous learning capacity for mitigating the catastrophic forgetting under cross-domain scenarios.

cs.CV

CL-BioGAN: Biologically-Inspired Cross-Domain Continual Learning for Hyperspectral Anomaly Detection

Memory stability and learning flexibility in continual learning (CL) is a core challenge for cross-scene Hyperspectral Anomaly Detection (HAD) task. Biological neural networks can actively forget history knowledge that conflicts with the learning of new experiences by regulating learning-triggered synaptic expansion and synaptic convergence. Inspired by this phenomenon, we propose a novel Biologically-Inspired Continual Learning Generative Adversarial Network (CL-BioGAN) for augmenting continuous distribution fitting ability for cross-domain HAD task, where Continual Learning Bio-inspired Loss (CL-Bio Loss) and self-attention Generative Adversarial Network (BioGAN) are incorporated to realize forgetting history knowledge as well as involving replay strategy in the proposed BioGAN. Specifically, a novel Bio-Inspired Loss composed with an Active Forgetting Loss (AF Loss) and a CL loss is designed to realize parameters releasing and enhancing between new task and history tasks from a Bayesian perspective. Meanwhile, BioGAN loss with L2-Norm enhances self-attention (SA) to further balance the stability and flexibility for better fitting background distribution for open scenario HAD (OHAD) tasks. Experiment results underscore that the proposed CL-BioGAN can achieve more robust and satisfying accuracy for cross-domain HAD with fewer parameters and computation cost. This dual contribution not only elevates CL performance but also offers new insights into neural adaptation mechanisms in OHAD task.

cs.CV

A Unified Data Augmentation Framework for Low-Resource Multi-Domain Dialogue Generation

Current state-of-the-art dialogue systems heavily rely on extensive training datasets. However, challenges arise in domains where domain-specific training datasets are insufficient or entirely absent. To tackle this challenge, we propose a novel data \textbf{A}ugmentation framework for \textbf{M}ulti-\textbf{D}omain \textbf{D}ialogue \textbf{G}eneration, referred to as \textbf{AMD$^2$G}. The AMD$^2$G framework consists of a data augmentation process and a two-stage training approach: domain-agnostic training and domain adaptation training. We posit that domain corpora are a blend of domain-agnostic and domain-specific features, with certain representation patterns shared among diverse domains. Domain-agnostic training aims to enable models to learn these common expressive patterns. To construct domain-agnostic dialogue corpora, we employ a \textit{\textbf{de-domaining}} data processing technique used to remove domain-specific features. By mitigating the effects of domain-specific features, the model trained on the de-domained corpora can effectively learn common expression patterns in different domains. Subsequently, we adapt the learned domain-agnostic features to the target domain through domain adaptation training. We conduct experiments on Chinese dialogue datasets from five different domains and show that AMD$^2$G achieves superior performance compared to both direct training on the target domain corpus and collective training on all five domain corpora. Our work underscores AMD$^2$G as a viable alternative solution for low-resource multi-domain dialogue generation. Code and data associated with our work are available on GitHub repository$^{\text 1}$.

cs.CL

Elliptic bihamiltonian structures from relative shifted Poisson structures

In this paper, generalizing the construction of \cite{HP1}, we equip the relative moduli stack of complexes over a Calabi-Yau fibration (possibly with singular fibers) with a shifted Poisson structure. Applying this construction to the anticanonical linear systems on surfaces, we get examples of compatible Poisson brackets on projective spaces extending Feigin-Odesskii Poisson brackets. Computing explicitly the corresponding compatible brackets coming from Hirzebruch surfaces, we recover the brackets defined by Odesskii-Wolf in \cite{OW}.

math.AG

Bosonization of Feigin-Odesskii Poisson varieties

The derived moduli stack of complexes of vector bundles on a Gorenstein Calabi-Yau curve admits a 0-shifted Poisson structure. Feigin-Odesskii Poisson varieties are examples of such moduli spaces over complex elliptic curves. Using moduli stack of chains we construct an auxiliary Poisson varieties with a Poisson morphism from it to a Feigin-Odesskii variety. We call it the \emph{bosonization} of Feigin-Odesskii variety. As an application, we show that the Feigin-Odesskii Poisson brackets on projective spaces (associated with stable bundles of arbitrary rank on elliptic curves) admit no infinitesimal symmetries.

math.AG

Cluster categories and rational curves

We study rational curves on smooth complex Calabi--Yau threefolds via noncommutative algebra. By the general theory of derived noncommutative deformations due to Efimov, Lunts and Orlov, the structure sheaf of a rational curve in a smooth CY 3-fold $Y$ is pro-represented by a nonpositively graded dg algebra $Γ$. The curve is called nc rigid if $H^0Γ$ is finite dimensional. When $C$ is contractible, $H^0Γ$ is isomorphic to the contraction algebra defined by Donovan and Wemyss. More generally, one can show that there exists a $Γ$ pro-representing the (derived) multi-pointed deformation (defined by Kawamata) of a collection of rational curves $C_1,\ldots,C_t$ so that ${\mathrm{dim}}({\rm{Hom}}_Y({\mathcal{O}}_{C_i},{\mathcal{O}}_{C_j}))=δ_{ij}$. The collection is called nc rigid if $H^0Γ$ is finite dimensional. We prove that $Γ$ is a homologically smooth bimodule 3CY algebra. As a consequence, we define a (2CY) cluster category ${\mathcal{C}}_Γ$ for such a collection of rational curves in $Y$. It has finite-dimensional morphism spaces iff the collection is nc rigid. When $\bigcup_{i=1}^tC_i$ is (formally) contractible by a morphism $\hat{Y}\to \hat{X}$, ${\mathcal{C}}_Γ$ is equivalent to the singularity category of $\hat{X}$ and thus categorifies the contraction algebra of Donovan and Wemyss. The Calabi-Yau structure on $Y$ determines a canonical class $[w]$ (defined up to right equivalence) in the zeroth Hochschild homology of $H^0Γ$. Using our previous work on the noncommutative Mather--Yau theorem and singular Hochschild cohomology, we prove that the singularities underlying a 3-dimensional smooth flopping contraction are classified by the derived equivalence class of the pair $(H^0Γ, [w])$. We also give a new necessary condition for contractibility of rational curves in terms of $Γ$.

math.AG

Orientation data on moduli space of sheaves on Calabi-Yau threefold

Kontsevich and Soibelman introduced a notion of orientation data on Calabi-Yau category. It can be viewed as a consistent choice of spin structure on moduli space of objects in the given category. The orientation data plays an important role in Donaldson-Thomas theory. Let X be a projective, simply connected and torsion free CY 3-fold. We prove the existence of orientation data for the moduli space of coherent sheaves on X.

math.AG

Quivers with analytic potentials

Given a quiver $Q$, a formal potential is called analytic if its coefficients are bounded by the terms of a geometric series. As shown by Toda, the potentials appearing in the deformation theory of complexes of coherent sheaves on complex projective Calabi-Yau threefolds are analytic. Our paper consists of two parts. In the first part, we establish the foundations of the differential calculus of quivers with analytic potentials and prove two fundamental results: the inverse function theorem and Moser's trick. As an application, we prove finite determinacy of analytic potentials with finite-dimensional Jacobi algebra, answering a question of Ben Davison. We also prove a Mather-Yau type theorem for analytic potentials with finite-dimensional Jacobi algebra, extending previous work by the first author with Zhou. In the second part, we study Donaldson-Thomas theory of quivers with analytic potentials. First, we construct a canonical perverse sheaf of vanishing cycles on the moduli stack of finite dimensional modules over the Jacobi algebra of an arbitrary iterated mutation via a separation lemma for analytic potentials. Finally, we provide a transformation formula for DT invariants (weighted by the Behrend function) under iterated mutations as a counterpart of Nagao's result on topological DT invariants. Our result leads to a perverse analogue of the $F$-series in cluster algebras. It also yields a Behrend--weighted Caldero--Chapoton formula for what we call anti-cluster algebras.

math.AG

Noncommutative Mather-Yau theorem and its applications to Calabi-Yau algebras

In this article, we prove that for a finite quiver $Q$ the equivalence class of a potential up to formal change of variables of the complete path algebra $\widehat{\mathbb{C} Q}$, is determined by its Jacobi algebra together with the class in its 0-th Hochschild homology represented by the potential assuming the Jacobi algebra is finite dimensional. This is an noncommutative analogue of the famous theorem of Mather and Yau on isolated hypersurface singularities. We also prove that the right equivalence class of a potential is determined by its sufficiently high jet assuming the Jacobi algebra is finite dimensional. These two theorems can be viewed as a first step towards the singularity theory of noncommutative power series. As an application, we show that if the Jacobi algebra is finite dimensional then the corresponding complete Ginzburg dg-algebra, and the (topological) generalized cluster category thereof, are determined by the isomorphic type of the Jacobi algebra together with the class in its 0-th Hochschild homology represented by the potential.

math.AG

Shifted Poisson structures and moduli spaces of complexes

In this paper we study the moduli stack of complexes of vector bundles (with chain isomorphisms) over a smooth projective variety $X$ via derived algebraic geometry. We prove that if $X$ is a Calabi-Yau variety of dimension $d$ then this moduli stack has a $(1-d)$-shifted Poisson structure. In the case $d=1$, we construct a natural foliation of the moduli stack by $0$-shifted symplectic substacks. We show that our construction recovers various known Poisson structures associated to complex elliptic curves, including the Poisson structure on Hilbert scheme of points on elliptic quantum projective planes studied by Nevins and Stafford, and the Poisson structures on the moduli spaces of stable triples over an elliptic curves considered by one of us. We also relate the latter Poisson structures to the semi-classical limits of the elliptic Sklyanin algebras studied by Feigin and Odesskii.

math.AG

Quasi-homogeneity of superpotentials

In this article, we study the quasi-homogeneity of a superpotential in a complete free algebra over an algebraic closed field of characteristic zero. We prove that a superpotential with finite dimensional Jacobi algebra is right equivalent to a weighted homogeneous superpotential if and only if the corresponding class in the 0-th Hochschlid homology group of the Jacobi algebra is zero. This result can be viewed as a noncommutative version of the famous theorem of Kyoji Saito on isolated hypersurface singularities.

math.AG

Contraction algebra and singularity of three-dimensional flopping contraction

In [5], Donovan and Wemyss introduced the contraction algebra of flopping curves in 3-folds. They conjectured that the contraction algebra determines the formal neighborhood of the underlying singularity of the contraction. In this paper, we prove that the contraction algebra together with its natural $A_\infty$-structure constructed in [10], determine the formal neighborhood of the singularity.

math.AG

Shifted Poisson geometry and meromorphic matrix algebras over an elliptic curve

In this paper we classify symplectic leaves of the regular part of the projectivization of the space of meromorphic endomorphisms of a stable vector bundle on an elliptic curve, using the study of shifted Poisson structures on the moduli of complexes from our previous work \cite{HP17}. This Poisson ind-scheme is closely related to the ind Poisson-Lie group associated to Belavin's elliptic $r$-matrix, studied by Sklyanin, Cherednik and Reyman and Semenov-Tian-Shansky. Our result leads to a classification of symplectic leaves on the regular part of meromorphic matrix algebras over an elliptic curve, which can be viewed as the Lie algebra of the above-mentioned ind Poisson-Lie group. We also describe the decomposition of the product of leaves under the multiplication morphism and show the invariance of Poisson structures under autoequivalences of the derived category of coherent sheaves on an elliptic curve.

math.AG

Contraction algebra and invariants of singularities

In [7], Donovan and Wemyss introduced the contraction algebra of flop- ping curves in 3-folds. When the flopping curve is smooth and irreducible, we prove that the contraction algebra together with its A_\infty-structure recovers various invariants associated to the underlying singularity and its small resolution, including the derived category of singularities and the genus zero Gopakuma-Vafa invariants.

math.AG

Chern-Simons functions on toric Calabi-Yau threefolds and Donaldson-Thomas theory

In this paper, we give a construction of the global Chern-Simons functions for toric Calabi-Yau stacks of dimension three using strong exceptional collections. The moduli spaces of sheaves on such stacks can be identified with critical loci of these functions. We give two applications of these functions. First, we prove Joyce's integrality conjecture of generalized DT invariants on local surfaces. Second, we prove a dimension reduction formula for virtual motives, which leads to two recursion formulas for motivic Donaldson-Thomas invariants.

math.AG

Classification of free actions on complete intersections of four quadrics

In this paper we classify all free actions of finite groups on Calabi-Yau complete intersection of 4 quadrics in $\PP^7$, up to projective equivalence. We get some examples of smooth Calabi-Yau threefolds with large nonabelian fundamental groups. We also observe the relation between some of these examples and moduli of polarized abelian surfaces.

math.AG