arXiv subjects

Xun Cai

Publications and source records attributed to Xun Cai.

13 recordsLinked to original sources

arXiv2026

Adaptive Depth-converted-Scale Convolution for Self-supervised Monocular Depth Estimation

Self-supervised monocular depth estimation (MDE) has received increasing interests in the last few years. The objects in the scene, including the object size and relationship among different objects, are the main clues to extract the scene structure. However, previous works lack the explicit handling of the changing sizes of the object due to the change of its depth. Especially in a monocular video, the size of the same object is continuously changed, resulting in size and depth ambiguity. To address this problem, we propose a Depth-converted-Scale Convolution (DcSConv) enhanced monocular depth estimation framework, by incorporating the prior relationship between the object depth and object scale to extract features from appropriate scales of the convolution receptive field. The proposed DcSConv focuses on the adaptive scale of the convolution filter instead of the local deformation of its shape. It establishes that the scale of the convolution filter matters no less (or even more in the evaluated task) than its local deformation. Moreover, a Depth-converted-Scale aware Fusion (DcS-F) is developed to adaptively fuse the DcSConv features and the conventional convolution features. Our DcSConv enhanced monocular depth estimation framework can be applied on top of existing CNN based methods as a plug-and-play module to enhance the conventional convolution block. Extensive experiments with different baselines have been conducted on the KITTI benchmark and our method achieves the best results with an improvement up to 11.6% in terms of SqRel reduction. Ablation study also validates the effectiveness of each proposed module.

cs.CV↗

arXiv2025

Seesaw of Saltwater and Inundation Drives Methane Emissions in Coastal Tidal Wetlands

Wetlands are significant carbon sinks, yet methane emissions partially offset this function due to its high global warming potential. Coastal tidal wetlands, unlike non-tidal wetlands, are regulated by oceanic drivers like salinity gradients and tidal inundation, which strongly influence methane production and release but remain poorly represented in regional assessments. Here, we estimate methane emissions from U.S. East Coast tidal marshes, by integrating ocean model, remote sensing datasets, empirical relationships from metadata. Spatially, emissions reflect the combined effects of marsh extent and per-unit-area flux rates, with hotspots occurring under lower salinity, higher inundation, and lower latitudes. Temporally, temperature and salinity dominate decadal-scale interannual variability. Between 2001 to 2020, total methane emissions are estimated at 0.019 - 0.038 Tg yr-1, with local fluxes rate ranging from 0 to 20 g m-2 day-1. Following pronounced hydrological variability in the early 2000s, emissions have increased steadily since 2007 at approximately 802 t yr-1, driven by warming, freshening, and enhanced inundation. Projections under IPCC climate scenarios indicate that increasing inundation will amplify methane emissions with sea-level rise, until a threshold near 0.75 m SLR, beyond which saltwater intrusion increasingly suppresses further growth, highlighting the critical role of salinity-inundation interactions in coastal methane dynamics.

physics.geo-ph↗

arXiv2025

Defending Saltwater Intrusion: The Freshwater Pushback

Saltwater Intrusion (SWI) threatens freshwater availability, agriculture, and ecosystem resilience in coastal regions. While sea-level rise (SLR) is a known driver of long-term salinization, the counteracting role of freshwater discharge remains underexamined. Here, we combine long-term observations with numerical modeling and machine learning reconstruction to quantify the buffering capacity of freshwater outflows across the U.S. coastline. In systems such as Delaware Bay and parts of the Gulf and South Atlantic coasts, the salt front has shifted seaward in recent decades, linked to increased discharge, despite SLR over that time period. We show that a 10 - 35% increase in freshwater flow can offset the salinity impact of 0.5 m of SLR, though regional variation is significant. With future discharge trends diverging spatially, SWI responses will be highly uneven. These results highlight the critical role of freshwater management in mitigating salinity risks under climate change, with implications for water resource resilience, coastal planning, and long-term adaptation strategies.

physics.geo-ph↗

arXiv2025

Recognizing Salt Wave Events in Coastal Systems

Saltwater intrusion is a critical challenge to coastal ecosystems, impacting freshwater resources, biogeochemical cycles, and habitat stability. While relevant studies often focus on the long-term trends of salinity, its episodic variability and resulting ecological disturbance remains underexplored. Here, we introduce the concept of "coastal salt waves" - periods of elevated salinity anomalies, akin to heat waves - to better characterize extreme salinity events and emphasize their significance. Using cases studies, we show that coastal salt waves exhibit significant spatial and temporal variability based on their duration, intensity, frequency, cumulation, and timing, with drivers and impacts that may differ from gradual saltwater intrusion. In years with similar average salinity, salt waves may still vary greatly in characteristics like intensity, resulting in varying environmental impacts. Furthermore, systems without rising average salinity may still face more frequent or intense salt waves. This framework supports monitoring and management strategies to mitigate coastal salinization risks.

physics.geo-ph↗

arXiv2025

MetricGrids: Arbitrary Nonlinear Approximation with Elementary Metric Grids based Implicit Neural Representation

This paper presents MetricGrids, a novel grid-based neural representation that combines elementary metric grids in various metric spaces to approximate complex nonlinear signals. While grid-based representations are widely adopted for their efficiency and scalability, the existing feature grids with linear indexing for continuous-space points can only provide degenerate linear latent space representations, and such representations cannot be adequately compensated to represent complex nonlinear signals by the following compact decoder. To address this problem while keeping the simplicity of a regular grid structure, our approach builds upon the standard grid-based paradigm by constructing multiple elementary metric grids as high-order terms to approximate complex nonlinearities, following the Taylor expansion principle. Furthermore, we enhance model compactness with hash encoding based on different sparsities of the grids to prevent detrimental hash collisions, and a high-order extrapolation decoder to reduce explicit grid storage requirements. experimental results on both 2D and 3D reconstructions demonstrate the superior fitting and rendering accuracy of the proposed method across diverse signal types, validating its robustness and generalizability. Code is available at https://github.com/wangshu31/MetricGrids}{https://github.com/wangshu31/MetricGrids.

cs.CV↗

arXiv2024

Approximately Invertible Neural Network for Learned Image Compression

Learned image compression have attracted considerable interests in recent years. It typically comprises an analysis transform, a synthesis transform, quantization and an entropy coding model. The analysis transform and synthesis transform are used to encode an image to latent feature and decode the quantized feature to reconstruct the image, and can be regarded as coupled transforms. However, the analysis transform and synthesis transform are designed independently in the existing methods, making them unreliable in high-quality image compression. Inspired by the invertible neural networks in generative modeling, invertible modules are used to construct the coupled analysis and synthesis transforms. Considering the noise introduced in the feature quantization invalidates the invertible process, this paper proposes an Approximately Invertible Neural Network (A-INN) framework for learned image compression. It formulates the rate-distortion optimization in lossy image compression when using INN with quantization, which differentiates from using INN for generative modelling. Generally speaking, A-INN can be used as the theoretical foundation for any INN based lossy compression method. Based on this formulation, A-INN with a progressive denoising module (PDM) is developed to effectively reduce the quantization noise in the decoding. Moreover, a Cascaded Feature Recovery Module (CFRM) is designed to learn high-dimensional feature recovery from low-dimensional ones to further reduce the noise in feature channel compression. In addition, a Frequency-enhanced Decomposition and Synthesis Module (FDSM) is developed by explicitly enhancing the high-frequency components in an image to address the loss of high-frequency information inherent in neural network based image compression. Extensive experiments demonstrate that the proposed A-INN outperforms the existing learned image compression methods.

eess.IV↗

arXiv2024

Quantum spin liquid from electron-phonon coupling

A quantum spin liquid (QSL) is an exotic insulating phase with emergent gauge fields and fractionalized excitations. However, the unambiguous demonstration of the existence of a QSL in a "non-engineered" microscopic model (or in any material) remains challenging. Here, using numerically-exact sign-problem-free quantum Monte Carlo simulations, we show that a QSL arises in a non-engineered electron-phonon model. Specifically, we investigate the ground-state phase diagram of the bond Su-Schrieffer-Heeger (SSH) model on a 2D triangular lattice at half filling (one electron per site) which we show includes a QSL phase which is fully gapped, exhibits no symmetry-breaking order, and supports deconfined fractionalized holon excitations. This suggests new routes for finding QSLs in realistic materials and high-$T_c$ superconductivity by lightly doping them.

cond-mat.str-el↗

arXiv2024

Towards Personalized Federated Multi-Scenario Multi-Task Recommendation

In modern recommender systems, especially in e-commerce, predicting multiple targets such as click-through rate (CTR) and post-view conversion rate (CTCVR) is common. Multi-task recommender systems are increasingly popular in both research and practice, as they leverage shared knowledge across diverse business scenarios to enhance performance. However, emerging real-world scenarios and data privacy concerns complicate the development of a unified multi-task recommendation model. In this paper, we propose PF-MSMTrec, a novel framework for personalized federated multi-scenario multi-task recommendation. In this framework, each scenario is assigned to a dedicated client utilizing the Multi-gate Mixture-of-Experts (MMoE) structure. To address the unique challenges of multiple optimization conflicts, we introduce a bottom-up joint learning mechanism. First, we design a parameter template to decouple the expert network parameters, distinguishing scenario-specific parameters as shared knowledge for federated parameter aggregation. Second, we implement personalized federated learning for each expert network during a federated communication round, using three modules: federated batch normalization, conflict coordination, and personalized aggregation. Finally, we conduct an additional round of personalized federated parameter aggregation on the task tower network to obtain prediction results for multiple tasks. Extensive experiments on two public datasets demonstrate that our proposed method outperforms state-of-the-art approaches. The source code and datasets will be released as open-source for public access.

cs.IR↗

arXiv2023

UNIDEAL: Curriculum Knowledge Distillation Federated Learning

Federated Learning (FL) has emerged as a promising approach to enable collaborative learning among multiple clients while preserving data privacy. However, cross-domain FL tasks, where clients possess data from different domains or distributions, remain a challenging problem due to the inherent heterogeneity. In this paper, we present UNIDEAL, a novel FL algorithm specifically designed to tackle the challenges of cross-domain scenarios and heterogeneous model architectures. The proposed method introduces Adjustable Teacher-Student Mutual Evaluation Curriculum Learning, which significantly enhances the effectiveness of knowledge distillation in FL settings. We conduct extensive experiments on various datasets, comparing UNIDEAL with state-of-the-art baselines. Our results demonstrate that UNIDEAL achieves superior performance in terms of both model accuracy and communication efficiency. Additionally, we provide a convergence analysis of the algorithm, showing a convergence rate of O(1/T) under non-convex conditions.

cs.LG↗

arXiv2023

High-temperature superconductivity induced by the Su-Schrieffer-Heeger electron-phonon coupling

Experimental quest for high-temperature and room-temperature superconductivity (SC) at ambient pressure has been a long-standing research theme in physics. It has also been desired to construct reliable microscopic mechanisms that may achieve high-temperature SC. Here we systematically explore SC in the Su-Schrieffer-Heeger (SSH) electron-phonon coupling models by performing numerically-exact quantum Monte-Carlo simulations. Our results reliably showed that superconducting $T_c$ of the SSH models is high, remarkably higher than those in the Holstein models, particularly in strong electron-phonon coupling regime. This is mainly because SSH phonons can not only induce strong pairing between electrons but also help the phase coherence of Cooper pairs, thus realizing higher $T_c$. As mechanism of higher-$T_c$ of the SSH models could be potentially relevant to realistic materials, it paves a promising way to find higher-temperature SC in the future.

cond-mat.str-el↗

arXiv2022

Completely Heterogeneous Federated Learning

Federated learning (FL) faces three major difficulties: cross-domain, heterogeneous models, and non-i.i.d. labels scenarios. Existing FL methods fail to handle the above three constraints at the same time, and the level of privacy protection needs to be lowered (e.g., the model architecture and data category distribution can be shared). In this work, we propose the challenging "completely heterogeneous" scenario in FL, which refers to that each client will not expose any private information including feature space, model architecture, and label distribution. We then devise an FL framework based on parameter decoupling and data-free knowledge distillation to solve the problem. Experiments show that our proposed method achieves high performance in completely heterogeneous scenarios where other approaches fail.

cs.LG↗

arXiv2021

Robustness of Antiferromagnetism in the Su-Schrieffer-Heeger-Hubbard model

We recently unveiled that antiferromagnetism (AFM) order can be dominantly induced by the bond Su-Schrieffer-Heeger (SSH) electron-phonon coupling (EPC), which is beyond the conventional wisdom that AFM order is usually driven by strong Coulomb interactions. Nevertheless, many aspects of the interplay between EPC and strong electronic interactions on AFM ordering remains unexplored. Here, we investigate the Su-Schrieffer-Heeger-Hubbard (SSHH) model with bond SSH phonons and onsite Hubbard interactions by large-scale quantum Monte-Carlo (QMC) simulations and obtain its ground-state phase diagram for various Hubbard interactions and EPC strength at half filling. Our results show that Hubbard interactions further enhance EPC-induced AFM, especially for small phonon frequency or in adiabatic limit, the regime most relevant to various quantum materials. This could shed further light to understanding the cooperative effect between EPC and electronic correlations in quantum materials.

cond-mat.str-el↗

arXiv2021

Antiferromagnetism induced by electron-phonon-coupling

Antiferromagnetism (AF) such as Neel ordering is often closely related to Coulomb interactions such as Hubbard repulsion in two-dimensional (2D) systems. Whether Neel AF ordering in 2D can be dominantly induced by electron-phonon couplings (EPC) has not been completely understood. Here, by employing numerically-exact sign-problem-free quantum Monte Carlo (QMC) simulations, we show that optical Su-Schrieffer-Heeger (SSH) phonons with frequency $ω$ and EPC constant $λ$ can induce AF ordering for a wide range of phonon frequency $ω>ω_c$. For $ω<ω_c$, a valence-bond-solid (VBS) order appears and there is a direct quantum phase transition between VBS and AF phases at $ω_c$. The phonon mechanism of the AF ordering is related to the fact that SSH phonons directly couple to electron hopping whose second-order process can induce an effective AF spin exchange. Our results shall shed new lights to understanding AF ordering in correlated quantum materials.

cond-mat.str-el↗