arXiv subjects

Hong Qin

Publications and source records attributed to Hong Qin.

At least 19 recordsLinked to original sources

arXiv2026

When Calibration Depends on the Scoring Rule: Quantized Biomedical LLM Classification

Quantized large language models can run on consumer hardware, which motivates interest in on-premises processing of sensitive data. The reliability of their confidence estimates depends on implementation choices (prompt template, label wording, and scoring normalization) that are seldom treated as experimental variables. We evaluate three 7-billion-parameter Mistral variants (a base model, BioMistral, and an instruction-tuned checkpoint evaluated without its chat template) at FP16, INT8, and INT4 on five-class sentence classification in medical abstracts. We test two primary templates on n = 2,000 test sentences and two auxiliary templates on n = 200 validation sentences. Our central observation, made without post-hoc calibration, is that switching from summed to mean-token log-likelihood reverses which model appears better calibrated: BioMistral's mean calibration error across matched conditions nearly triples, while the instruction-tuned model's drops by more than half. Accuracy changes by at most 1.4 percentage points for these two checkpoints. Negative log-likelihood and Brier score show the same reversal. The two primary templates were selected using test-derived examples, so absolute performance with them is exploratory. Between them, prompt choice changes mean accuracy across precisions by 2.9 to 17.8 percentage points. Eight-bit quantization changes accuracy by at most 1.1 percentage points for the adapted checkpoints; four-bit quantization shows mixed but non-catastrophic effects. Post-hoc temperature scaling reduces calibration error under summed scoring but was not fitted under mean-token scoring, so whether the reversal survives per-scorer calibration is unknown. These exploratory results suggest that calibration comparisons of decoder-based classifiers should treat scoring normalization and prompt design as first-order experimental decisions.

cs.LG↗

arXiv2026

Thresholdless IBW Emission and Alpha-to-Thermal-Ion Energy Channeling via Pole Resonance of Fusion-Product Ions

Fusion reactions can release a substantial fraction, and in aneutronic reactions nearly all, of their energy as the kinetic energy of fusion-product ions. In magnetized plasmas, these energetic ions can form ring distributions in velocity space. Such distributions can drive Dory-Guest-Harris (DGH) electrostatic instabilities, but these self-instabilities require a finite energetic-ion density and can be stabilized by a thermal background. We show that the same background can instead enable a distinct instability mechanism: a cyclotron pole of the minority energetic-ion susceptibility destabilizes a stable ion Bernstein wave (IBW) eigenmode. When the energetic-ion harmonic is distinct from the thermal-ion harmonics, exact root-pole resonance is thresholdless in the ideal collisionless limit. A species-resolved power balance shows that the energetic ions supply the free energy while the thermal plasma receives it. In the LAPD proton--alpha example, $99.96\%$ of the alpha-particle power loss enters the coherent proton response. This self-excited, ion-directed transfer provides a possible linear building block for alpha-particle energy channeling in proton-Boron11 fusion.

physics.plasm-ph↗

arXiv2026

Electrons quench ion tails during evaporative cooling in hot ion mode plasmas

It is shown how evaporative cooling in hot plasma is essentially different from evaporative cooling in other media, such as neutral gases or Bose-Einstein condensates. The fundamental difference in plasmas arises both from the large mass ratio between electrons and ions in fully ionized plasma and the unusually sensitive dependence of plasma collisionality on speed. Thus, a hot ion mode plasma ($T_e < T_i$) is shown to support a distinctive evaporative cooling regime, where a strong reduction in the ion evaporation rate appears. This new regime may have application to approaches to economical nuclear fusion, where the ion tail plays an outsized role in the fusion reaction rate.

physics.plasm-ph↗

arXiv2026

NeuraLSP: A Neural Spectral Preconditioner for Accelerating PDE Solvers

Solving large-scale sparse linear systems originating from partial differential equations (PDEs) is a fundamental topic in high-performance scientific computing, where preconditioners are crucial. Multigrid methods are among the most effective preconditioners, yet their performance is dictated by the accurate construction of grid transfer operators. Current neural multigrid methods learn such operators with graph neural networks (GNNs), typically by extracting connectivity from discretized system matrices. While effective, these graph-based constructions suffer from rank inflation, resulting in unnecessarily large coarse spaces and slower convergence. To ameliorate, this paper advocates NeuraLSP, a new neural multigrid preconditioner that replaces graph aggregation with a fixed low-rank spectral representation derived from the left singular subspace of near-nullspace components. At the network design level, NeuraLSP is trained with a novel subspace loss function, which preserves the error modes most relevant to multigrid convergence while suppressing rank inflation. This paper's grand innovation hinges upon both theoretical guarantees and empirical robustness to rank inflation, affording up to a 53% speedup over SOTA neural preconditioners across a variety of PDE families.

cs.LG↗

arXiv2026

Inferring and Predicting Clade-Level Relative Transmission Fitness in Seasonal Influenza A Using Differential Population Growth Rate and Deep Learning

Seasonal influenza A evolves rapidly, allowing newly emerged clades to replace previously dominant lineages and complicate surveillance and vaccine evaluation. Here, we applied the Differential Population Growth Rate (DPGR) framework to GISAID-derived H3N2 and H1N1 surveillance data collected from 1 January 2014 to 12 February 2026, including the 2025-2026 influenza season, to estimate clade-level relative transmission fitness across continents and within the United States. We identified windows of co-circulation with sliding-window regression, reconstructed relative-fitness relationships among clades, and compared inferred growth advantages with independent WHO and CDC surveillance patterns. We further trained subtype-specific convolutional neural networks on complete viral genomes to predict DPGR from sequence, quantified predictive uncertainty with conformal prediction, and used SHAP to localize genomic contributors to fitness. DPGR recovered recurrent lineage turnover in both subtypes and consistently identified the emerging H3N2 subclade K as fitter than the 2025-2026 vaccine-lineage background across multiple regions. Genome-based models predicted DPGR accurately for H3N2 ($R^2 = 0.9577$) and H1N1 ($R^2 = 0.9871$), while interpretation highlighted known haemagglutinin antigenic sites together with contributions from internal genes. These results support DPGR as an interpretable surveillance signal and show that influenza fitness can be linked to genomic prediction and biological interpretation in a unified framework.

q-bio.PE↗

arXiv2026

Differential privacy statistical inference for a directed graph network model with covariates

Network data typically contain sensitive relational information, where direct release or sharing may lead to non-negligible privacy violations without proper statistical safeguards. While differential privacy has emerged as a powerful framework for privacy-preserving network data analysis, theoretical understanding remains limited particularly for models incorporating both network structure and nodal attributes. This paper bridges this gap by investigating a directed $β$-model with covariates under differential privacy constraints. Our model accounts for both node-level heterogeneity (via $2n$-dimensional degree parameters $θ$ ) and covariate-driven homogeneity (via a $p$-dimensional parameter $γ$). To protect privacy, we introduce a joint Laplace mechanism for releasing network statistics while satisfying differential privacy constraints. Leveraging moment-based estimation techniques, we estimate the parameters of both degree heterogeneity and homogeneity and derive the consistency and asymptotic normality of the differentially private estimators as the network size tends to infinity. Our theoretical findings are validated through numerical simulations and real-world case studies, demonstrating the validity of our theoretical results.

math.ST↗

arXiv2026

On the Connection Between Differential Population Growth Rate and Epidemic Reproduction Numbers

During pandemics, public health agencies need to rapidly assess whether a new viral variant is more transmissible than existing lineages. For co-circulating variants, relative fitness can be expressed as a selective coefficient, as the differential population growth rate (DPGR) estimated from genomic surveillance, or, with additional assumptions, as a contrast in epidemic reproduction numbers $R_t$. We show that DPGR estimates a pairwise growth-rate difference. Under a specified generation-interval model, this difference can be transformed into reproduction-number space; in the equal-generation-time SIR special case, it reduces to a scaled difference in variant-specific $R_t$. Related growth-rate contrasts also appear in multinomial logistic and growth-advantage random-walk models, although those methods differ from DPGR in likelihood, smoothing, priors, and data inputs. We evaluate the theory across five SARS-CoV-2 and influenza analyses totaling more than 2,200 matched data points. SIR simulation recovers the expected mapping when the true $R_t$ is known, and retrospective SARS-CoV-2 analyses show sustained DPGR signals 43 to 65 days before variant dominance, with 95\% sign accuracy in our analysis. DPGR is approximately transitive across lineage triplets, near zero for selected functionally similar sublineages, and directionally consistent across countries. These results connect sequence-count-based fitness estimates to reproduction-number contrasts through an assumption-explicit growth-rate bridge.

q-bio.PE↗

arXiv2026

Warm Topological Langmuir Cyclotron Wave

Finite-temperature effects in magnetized electron plasmas create a new Weyl-point degeneracy between the warm Langmuir and right-circularly polarized waves. The associated topological charge at this warm Weyl point is found to be 1, which, by the index theorem, predicts a gap-traversing topological edge mode. Solving the full warm-fluid eigenmode problem In a 1D inhomogeneous equilibrium, we numerically identify this anticipated mode as the warm topological Langmuir-cyclotron wave, which is absent in the cold limit and occurs in a parameter regime relevant to the LArge Plasma Device (LAPD) at UCLA.

physics.plasm-ph↗

arXiv2026

Generalized Yee methods: Scalable symplectic finite element Maxwell solvers

Yee's finite-difference method preserves two crucial properties of Maxwell's equations -- locality and symplecticity -- and thereby enjoys two computational advantages: scalability on high-performance architectures and long-time numerical accuracy. In this work, we show that Yee's method is a special case of a class of structure-preserving finite element methods -- termed generalized Yee methods (GYMs) -- that are designed to retain both crucial properties. GYMs are built from de Rham-conforming finite elements and achieve locality through sparse mass matrices and their sparse approximate inverses (SPAIs). We prove that the symplectic structure of GYMs is invariant under such sparse approximations, freeing the choice of sparsification strategy. We introduce a novel sparsification strategy, SPAI-OP, which concentrates accuracy at prescribed wave modes by operator probing. We further extend GYMs to structure-preserving electromagnetic particle-in-cell (PIC) methods, whose symplecticity over particle trajectories requires the smooth fields afforded by higher-order finite elements. GYMs therefore retain the computational virtues of Yee's method while enabling unstructured meshes, higher-order accuracy, spectral adaptivity, and symplectic particle coupling.

math.NA↗

arXiv2026

Self-Consistent Dynamics of Electron Radiation Reaction via Structure-Preserving Geometric Algorithms for Coupled Schrödinger-Maxwell Systems

Classically, a charged particle in a magnetic field emits radiation, losing momentum and experiencing the Abraham-Lorentz (AL) / Landau-Lifshitz (LL) radiation reaction (RR) force. However, at atomic scales and outside the range of their applicability, the AL/LL equations fail and RR destroys the coherent state of an electron-undermining the very concept of a RR force. This process can be described by the coupled Schrödinger-Maxwell (SM) system under appropriate limits, but the system's nonlinear complexity has long limited purely analytical studies. We present geometric structure-preserving algorithms for the SM system that preserve gauge invariance, symplecticity, and unitarity on the discrete space-time lattice, which are implemented in our Structure-Preserving scHrodINger maXwell (SPHINX) code. By constructing coherent states from the Landau levels, SPHINX simulates the fully-coupled nonlinear dynamics of an electron coherent state, the energy partition evolution, and decoherence/relaxation of the electron wave packet in time due to RR. These simulations indicate that, in an external magnetic field, an electron prepared in an atomic-scale coherent state can radiate strongly, rapidly losing coherence and dispersing into a decoherent wave packet. Additionally, we also present the fully-coupled nonlinear evolution of the non-degenerate ground- and first-excited Landau levels themselves to understand how the coupled SM system modifies the well-known ideal (i.e., Schrödinger-only) dynamics of the Landau Levels. With appropriate boundary conditions, simulations show that the Landau levels are renormalized into stationary dressed eigenstates with constant electromagnetic and kinetic energies. This opens a new computational window into RR physics and advances modeling of extreme-field phenomena in fusion plasmas, astrophysics, and next-generation laser experiments

physics.plasm-ph↗

arXiv2026

2DMamba: Efficient State Space Model for Image Representation with Applications on Giga-Pixel Whole Slide Image Classification

Efficiently modeling large 2D contexts is essential for various fields including Giga-Pixel Whole Slide Imaging (WSI) and remote sensing. Transformer-based models offer high parallelism but face challenges due to their quadratic complexity for handling long sequences. Recently, Mamba introduced a selective State Space Model (SSM) with linear complexity and high parallelism, enabling effective and efficient modeling of wide context in 1D sequences. However, extending Mamba to vision tasks, which inherently involve 2D structures, results in spatial discrepancies due to the limitations of 1D sequence processing. On the other hand, current 2D SSMs inherently model 2D structures but they suffer from prohibitively slow computation due to the lack of efficient parallel algorithms. In this work, we propose 2DMamba, a novel 2D selective SSM framework that incorporates the 2D spatial structure of images into Mamba, with a highly optimized hardware-aware operator, adopting both spatial continuity and computational efficiency. We validate the versatility of our approach on both WSIs and natural images. Extensive experiments on 10 public datasets for WSI classification and survival analysis show that 2DMamba improves up to 2.48% in AUC, 3.11% in F1 score, 2.47% in accuracy and 5.52% in C-index. Additionally, integrating our method with VMamba for natural imaging yields 0.5 to 0.7 improvements in mIoU on the ADE20k semantic segmentation dataset, and 0.2% accuracy improvement on ImageNet-1K classification dataset. Our code is available at https://github.com/AtlasAnalyticsLab/2DMamba.

cs.CV↗

arXiv2026

Hidden free energy released by explicit parity-time-symmetry breaking

It is shown that the familiar two-stream instability is the result of spontaneous parity-time (PT)-symmetry breaking in a conservative system, and more importantly, explicit PT-symmetry breaking by viscosity can destabilize the system in certain parameter regimes that are stable when viscosity vanishes. This reveals that complex systems may possess hidden free energies protected by PT-symmetry and viscosity, albeit dissipative, can expose the systems to these freed energies by breaking PT-symmetry explicitly. Such a process is accompanied by instability and total variation growth.

physics.plasm-ph↗

arXiv2025

LBMamba: Locally Bi-directional Mamba

Mamba, a State Space Model (SSM) that accelerates training by recasting recurrence as a parallel scan, has recently emerged as a linearly-scaling alternative to self-attention. Because of its unidirectional nature, each state in Mamba only has information of its previous states and is blind to states after. Current Mamba-based computer-vision methods typically overcome this by augmenting Mamba's global forward scan with a global backward scan, forming a bi-directional scan to restore a full receptive field. However, this operation doubles the computational load, eroding much of the efficiency advantage that originally Mamba have. To eliminate this extra scans, we introduce LBMamba, a locally bi-directional SSM block that embeds a lightweight locally backward scan inside the forward scan and executes it in per-thread registers. Building on LBMamba, we present LBVim, a backbone that alternates scan directions every two layers to recover a global receptive field without extra backward sweeps. We validate our approach on both natural images and whole slide images (WSIs) and show that it constantly offers a superior performance-throughput trade-off. Under the same throughput, LBVim achieves 0.8% to 1.6% higher top-1 accuracy on the ImageNet-1K classification dataset, 0.6% to 2.7% higher mIoU on the ADE20K semantic segmentation dataset, 0.9% higher APb and 1.1% higher APm on the COCO detection dataset. Our method also boosts the accuracy of four SOTA Mamba models, namely VMamba, LocalVim, PlainMamba and Adventurer, by 0.5% to 3.4%. We integrate LBMamba into the SOTA pathology multiple instance learning (MIL) model, MambaMIL, which is unidirectional. Experiments on 3 public WSI classification datasets show that our method achieves a relative improvement of up to 3.06% better AUC, 3.39% better F1, 1.67% better accuracy. Our code is available at https://github.com/cvlab-stonybrook/LBMamba.

cs.CV↗

arXiv2025

MIND: Material Interface Generation from UDFs for Non-Manifold Surface Reconstruction

Unsigned distance fields (UDFs) are widely used in 3D deep learning due to their ability to represent shapes with arbitrary topology. While prior work has largely focused on learning UDFs from point clouds or multi-view images, extracting meshes from UDFs remains challenging, as the learned fields rarely attain exact zero distances. A common workaround is to reconstruct signed distance fields (SDFs) locally from UDFs to enable surface extraction via Marching Cubes. However, this often introduces topological artifacts such as holes or spurious components. Moreover, local SDFs are inherently incapable of representing non-manifold geometry, leading to complete failure in such cases. To address this gap, we propose MIND (Material Interface from Non-manifold Distance fields), a novel algorithm for generating material interfaces directly from UDFs, enabling non-manifold mesh extraction from a global perspective. The core of our method lies in deriving a meaningful spatial partitioning from the UDF, where the target surface emerges as the interface between distinct regions. We begin by computing a two-signed local field to distinguish the two sides of manifold patches, and then extend this to a multi-labeled global field capable of separating all sides of a non-manifold structure. By combining this multi-labeled field with the input UDF, we construct material interfaces that support non-manifold mesh extraction via a multi-labeled Marching Cubes algorithm. Extensive experiments on UDFs generated from diverse data sources, including point cloud reconstruction, multi-view reconstruction, and medial axis transforms, demonstrate that our approach robustly handles complex non-manifold surfaces and significantly outperforms existing methods. The source code is available at https://github.com/jjjkkyz/MIND.

cs.CV↗

arXiv2025

Wave Topology in Hall MHD

Hall Magnetohydrodynamics (HMHD) extends ideal MHD by incorporating the Hall effect via the induction equation, making it more accurate for describing plasma behavior at length scales below the ion skin depth. Despite its importance, a comprehensive description of the eigenmodes in HMHD has been lacking. In this work, we derive the complete spectrum and eigenvectors of HMHD waves and identify their underlying topological structure. We prove that the HMHD wave spectrum is homotopic to that of ideal MHD, consisting of three distinct branches: the slow magnetosonic-Hall waves, the shear Alfvén-Hall waves, and the fast magnetosonic-Hall waves, which continuously reduce to their ideal MHD counterparts in the limit of vanishing Hall parameter. Contrary to a recent claim, we find that HMHD does not admit any additional wave branches beyond those in ideal MHD. The key qualitative difference lies in the topological nature of the HMHD wave structure: it exhibits nontrivial topology characterized by a Weyl point-an isolated eigenmode degeneracy point-and associated nonzero Chern numbers of the eigenmode bundles over a 2-sphere in k-space surrounding the Weyl point.

physics.plasm-ph↗

arXiv2025

Sliced Space-filling Design with Mixtures

In this paper, we proposes the construction methods of sliced space-filling design when the quantitative factors are mixture components. Leveraging the representative points framework for distribution and energy distance decomposition theory, this paper proposes three methods for constructing sliced representative points and establishes their distributional convergence. Furthermore, one-shot and sequential algorithms for generating sliced space-filling mixture design for experiments with process variables are presented with convergence proofs. Compared to existing methods, the proposed sliced space-filling mixture design exhibits greater flexibility in subdesign run sizes and broader applicability to constrained experimental regions. Moreover, numerical results confirm its marked advantages in both space-filling performance and predictive accuracy.

math.ST↗

arXiv2025

Quantum Inspiration, Classical Advantage: Dequantized particle algorithm for the nonlinear Vlasov-Poisson system

We present a dequantization algorithm for the Vlasov--Poisson (VP) system, termed the dequantized particle algorithm, by systematically dequantizing the underlying many-body quantum theory. Starting from the second-quantized Hamiltonian description, we derive a finite-dimensional dequantized system and show that it furnishes a structure-preserving discretization of the Schrödinger--Poisson (SP) equations. Through the Wigner or Husimi transformations, this discretization provides an efficient approximation of the VP system when quantum effects are negligible. Unlike conventional structure-preserving algorithms formulated in 6D phase space, this dequantized particle algorithm operates in 3D configuration space, potentially offering more compact and efficient representations of physical information under appropriate conditions. A numerical example of the classical nonlinear two-stream instability, simulated using merely 97 dequantized particles, demonstrates the efficiency, accuracy, and conservation properties of the algorithm and confirms its potential as a foundation for developing quantum and quantum-inspired classical algorithms for kinetic plasma dynamics.

physics.plasm-ph↗

arXiv2025

Closed-form Solutions: A New Perspective on Solving Differential Equations

The quest for analytical solutions to differential equations has traditionally been constrained by the need for extensive mathematical expertise. Machine learning methods like genetic algorithms have shown promise in this domain, but are hindered by significant computational time and the complexity of their derived solutions. This paper introduces SSDE (Symbolic Solver for Differential Equations), a novel reinforcement learning-based approach that derives symbolic closed-form solutions for various differential equations. Evaluations across a diverse set of ordinary and partial differential equations demonstrate that SSDE outperforms existing machine learning methods, delivering superior accuracy and efficiency in obtaining analytical solutions.

cs.LG↗