arXiv subjects

Zhenyuan Wang

Publications and source records attributed to Zhenyuan Wang.

10 recordsLinked to original sources

arXiv2026

Lossless Compression of Cosmological Information from Type Ia Supernova Distance Measurements

We perform model-independent distance measurements on four Type Ia supernovae (SNe Ia) compilations (Pantheon, Pantheon+, DES-Dovekie, Union3) and compress each dataset into the values of $\log r_p(z)$ at eleven redshift knots, where $r_p(z)$ is a rescaled comoving distance. These Gaussian distributed compressed values, together with their full covariance, completely capture the distance-redshift relation information from each dataset. We demonstrate this by using these to perform an Markov Chain Monte Carlo (MCMC) likelihood analysis to infer cosmological parameters in flat $Λ$CDM, flat $w_0 w_a$CDM, and a non-parametric reconstruction of the dark-energy density $X(z) \equiv ρ_{\rm DE}(z)/ρ_{\rm DE}(0)$. Across all datasets and flux-averaging configurations and all three cosmological models, the resulting parameter contours and figures of merit reproduce the corresponding full distance-modulus analyses using the original SNe Ia data sets within the statistical sampling noise of the chains, demonstrating that the eleven $\log r_p$ data points are an operationally lossless compression of the cosmological information in the dataset. Our SN Ia data compression enables an analytic analysis that completes in $O(10^{-2})$ s per dataset and reduces the downstream cosmological MCMC to the fast evaluation of an $11$-dimensional Gaussian likelihood, with a per-step cost set by the number of knots and independent of the SNe Ia sample size. Our methodology will benefit the data analysis of future surveys from Euclid, Roman, and LSST, which will deliver SNe Ia samples one to three orders of magnitude larger than current ones.

astro-ph.CO↗

arXiv2026

Model-Independent Analysis of Type Ia Supernova Datasets and Implications for Dark Energy

Recent analyses combining DESI DR2 BAO with CMB and SNe Ia data have reported $2.8$--$4.2σ$ evidence for dynamical dark energy, but the significance depends strongly on the supernova sample, raising the question of whether this signal reflects new physics, dataset-specific systematics, or the choice of dark energy parameterization. We investigate this question by analyzing four SNe Ia compilations (Pantheon, Pantheon+, DES-Dovekie, and Union3) with DESI DR2 BAO and Planck CMB distance priors, using flux averaging, model-independent expansion rate extraction, parametric ($w_0 w_a$CDM) fits, and a non-parametric reconstruction of the dark energy density ratio $X(z) \equiv ρ_{\rm DE}(z)/ρ_{\rm DE}(0)$. Flux averaging reduces the $Ω_m$ difference between SNe and DESI from ${\sim}2σ$ to ${\sim}1σ$ for Pantheon+ and DES-Dovekie. The reconstructed $X(z)$ for DESI DR2 + CMB + SNe is consistent with $Λ$CDM for Pantheon, Pantheon+, and DES-Dovekie except at $0.5<z<1$, consistent with Wang \& Freese (2026). The largest deviation occurs at $z=2/3$, reaching ${\sim}2.7σ$ for Pantheon+ but only $1.6$--$1.7σ$ for Pantheon and DES-Dovekie. The $X(z)$ for DESI DR2 + CMB + Union3 is consistent with these within $1σ$, but shows an additional $2.4σ$ deviation at $z=1/3$ besides the ${\sim}2.7σ$ deviation at $z=2/3$. Across all analyses, the departure from $Λ$CDM correlates with each dataset's $Ω_m$ preference. We demonstrate that a pure $Λ$CDM universe with the measured $Ω_m$ differences can reproduce the observed $X(z)$ pattern, providing a viable alternative interpretation of the observed $X(z) \neq 1$ pattern. Future surveys by Euclid and Roman with sub-percent $Ω_m$ constraints will be essential to determine whether the signal reflects genuine dark energy evolution or residual inter-probe $Ω_m$ inconsistencies.

astro-ph.CO↗

arXiv2025

FEAT: A Multi-Agent Forensic AI System with Domain-Adapted Large Language Model for Automated Cause-of-Death Analysis

Forensic cause-of-death determination faces systemic challenges, including workforce shortages and diagnostic variability, particularly in high-volume systems like China's medicolegal infrastructure. We introduce FEAT (ForEnsic AgenT), a multi-agent AI framework that automates and standardizes death investigations through a domain-adapted large language model. FEAT's application-oriented architecture integrates: (i) a central Planner for task decomposition, (ii) specialized Local Solvers for evidence analysis, (iii) a Memory & Reflection module for iterative refinement, and (iv) a Global Solver for conclusion synthesis. The system employs tool-augmented reasoning, hierarchical retrieval-augmented generation, forensic-tuned LLMs, and human-in-the-loop feedback to ensure legal and medical validity. In evaluations across diverse Chinese case cohorts, FEAT outperformed state-of-the-art AI systems in both long-form autopsy analyses and concise cause-of-death conclusions. It demonstrated robust generalization across six geographic regions and achieved high expert concordance in blinded validations. Senior pathologists validated FEAT's outputs as comparable to those of human experts, with improved detection of subtle evidentiary nuances. To our knowledge, FEAT is the first LLM-based AI agent system dedicated to forensic medicine, offering scalable, consistent death certification while maintaining expert-level rigor. By integrating AI efficiency with human oversight, this work could advance equitable access to reliable medicolegal services while addressing critical capacity constraints in forensic systems.

cs.AI↗

arXiv2025

MERE: Hardware-Software Co-Design for Masking Cache Miss Latency in Embedded Processors

Runahead execution is a technique to mask memory latency caused by irregular memory accesses. By pre-executing the application code during occurrences of long-latency operations and prefetching anticipated cache-missed data into the cache hierarchy, runahead effectively masks memory latency for subsequent cache misses and achieves high prefetching accuracy; however, this technique has been limited to superscalar out-of-order and superscalar in-order cores. For implementation in scalar in-order cores, the challenges of area-/energy-constraint and severe cache contention remain. Here, we build the first full-stack system featuring runahead, MERE, from SoC and a dedicated ISA to the OS and programming model. Through this deployment, we show that enabling runahead in scalar in-order cores is possible, with minimal area and power overheads, while still achieving high performance. By re-constructing the sequential runahead employing a hardware/software co-design approach, the system can be implemented on a mature processor and SoC. Building on this, an adaptive runahead mechanism is proposed to mitigate the severe cache contention in scalar in-order cores. Combining this, we provide a comprehensive solution for embedded processors managing irregular workloads. Our evaluation demonstrates that the proposed MERE attains 93.5% of a 2-wide out-of-order core's performance while constraining area and power overheads below 5%, with the adaptive runahead mechanism delivering an additional 20.1% performance gain through mitigating the severe cache contention issues.

cs.AR↗

arXiv2024

Perturbation Theory Remixed II: Improved Modeling of Nonlinear Bispectrum

We present the application of the $n$-th order Eulerian Perturbation Theory ($n$EPT) for modeling the matter bispectrum in real space as an advancement over the Standard Perturbation Theory (SPT). The $n$EPT method, detailed in Wang et al. (2023) \cite{Wang2023nEPT}, sums up the density perturbations up to the $n$-th order before computing summary statistics such as bispectrum. Taking advantage of grid-based calculation of SPT (GridSPT), we make a realization-based comparison of the analytical nonlinear bispectrum predictions from $n$EPT and SPT against a suite of $N$-body simulations. Using a spherical-bispectrum visualization scheme, we show that $n$EPT bispectrum matches better than SPT bispectrum over a wide range of scales in general $w$CDM cosmologies. Like the power spectrum case, we find that $n$EPT bispectrum modeling accuracy is controlled by $σ_8(z) \equiv σ_8 D(z)$, where $D(z)$ is the linear growth factor at a redshift $z$. Notably, the 6EPT doubles the bispectrum model's validity range compared to the one-loop SPT for $σ_8(z) < 0.5$, corresponding to redshifts $z\ge1$ for the best-fitting Planck-2018 cosmology. For $n\ge5$, however, $n$EPT bispectrum depends sensitively on the cut-off scale or the grid resolution. The percent-level modeling accuracy achieved for the spherical bispectrum (where we average over all triangular configurations) becomes much degraded when fixing configurations. Thus, we show that the validity range of the field-level cosmological inferences must be different from that derived from averaged summary statistics such as $n$-point correlation functions.

astro-ph.CO↗

arXiv2024

Large-vocabulary forensic pathological analyses via prototypical cross-modal contrastive learning

Forensic pathology is critical in determining the cause and manner of death through post-mortem examinations, both macroscopic and microscopic. The field, however, grapples with issues such as outcome variability, laborious processes, and a scarcity of trained professionals. This paper presents SongCi, an innovative visual-language model (VLM) designed specifically for forensic pathology. SongCi utilizes advanced prototypical cross-modal self-supervised contrastive learning to enhance the accuracy, efficiency, and generalizability of forensic analyses. It was pre-trained and evaluated on a comprehensive multi-center dataset, which includes over 16 million high-resolution image patches, 2,228 vision-language pairs of post-mortem whole slide images (WSIs), and corresponding gross key findings, along with 471 distinct diagnostic outcomes. Our findings indicate that SongCi surpasses existing multi-modal AI models in many forensic pathology tasks, performs comparably to experienced forensic pathologists and significantly better than less experienced ones, and provides detailed multi-modal explainability, offering critical assistance in forensic investigations. To the best of our knowledge, SongCi is the first VLM specifically developed for forensic pathological analysis and the first large-vocabulary computational pathology (CPath) model that directly processes gigapixel WSIs in forensic science.

eess.IV↗

arXiv2024

Reconstructing the Long-wavelength Matter Density Fluctuation Modes from the Scalar-Type Clustering Fossils

Revealing the large-scale structure from the 21cm intensity mapping surveys is only possible after the foreground cleaning. However, most current cleaning techniques relying on the smoothness of the foreground spectrum lead to a severe side effect of removing the large-scale structure signal along the line of sight. On the other hand, the clustering fossil, a coherent variation of the small-scale clustering over large scales, allows us to recover the long-wavelength density modes from the off-diagonal correlation between short-wavelength modes. In this paper, we study the requirements for an unbiased and optimal clustering-fossil estimator and show that (A) the estimator is unbiased only when using an accurate bispectrum model for the long-short-short mode coupling and (B) including the connected four-point correlation functions is essential for characterizing the noise power spectrum of the estimated long mode. The clustering fossil estimator based upon the leading-order bispectrum yields an unbiased estimation of the long-wavelength ($k\lesssim 0.01~[h/{\rm Mpc}]$) modes with the cross-correlation coefficient of $0.7$ at redshifts $z=0$ to $3$.

astro-ph.CO↗

arXiv2023

Forensic Histopathological Recognition via a Context-Aware MIL Network Powered by Self-Supervised Contrastive Learning

Forensic pathology is critical in analyzing death manner and time from the microscopic aspect to assist in the establishment of reliable factual bases for criminal investigation. In practice, even the manual differentiation between different postmortem organ tissues is challenging and relies on expertise, considering that changes like putrefaction and autolysis could significantly change typical histopathological appearance. Developing AI-based computational pathology techniques to assist forensic pathologists is practically meaningful, which requires reliable discriminative representation learning to capture tissues' fine-grained postmortem patterns. To this end, we propose a framework called FPath, in which a dedicated self-supervised contrastive learning strategy and a context-aware multiple-instance learning (MIL) block are designed to learn discriminative representations from postmortem histopathological images acquired at varying magnification scales. Our self-supervised learning step leverages multiple complementary contrastive losses and regularization terms to train a double-tier backbone for fine-grained and informative patch/instance embedding. Thereafter, the context-aware MIL adaptively distills from the local instances a holistic bag/image-level representation for the recognition task. On a large-scale database of $19,607$ experimental rat postmortem images and $3,378$ real-world human decedent images, our FPath led to state-of-the-art accuracy and promising cross-domain generalization in recognizing seven different postmortem tissues. The source code will be released on \href{https://github.com/ladderlab-xjtu/forensic_pathology}{https://github.com/ladderlab-xjtu/forensic\_pathology}.

cs.CV↗

arXiv2022

Perturbation Theory Remixed: Improved Nonlinearity Modeling beyond Standard Perturbation Theory

We present a novel $n$EPT ($n$th-order Eulerian Perturbation Theory) scheme to model the nonlinear density field by the summation up to $n$th-order density fields in perturbation theory. The obtained analytical power spectrum shows excellent agreement with the results from all 20 Dark-Quest suites of $N$-body simulations spreading over a broad range of cosmologies. The agreement is much better than the conventional two-loop Standard Perturbation Theory and would reach out to $k_{\rm max}\simeq 0.4~h/{\rm Mpc}$ at $z=3$ for the best-fitting Planck cosmology, without any free parameters. The method can accelerate the forward modeling of the non-linear cosmological density field, an indispensable probe of cosmic mysteries such as inflation, dark energy, and dark matter.

astro-ph.CO↗

arXiv2020

The Breakdown Scale of HI Bias Linearity

The 21 cm intensity mapping experiments promise to obtain the large-scale distribution of HI gas at the post-reionization epoch. In order to reveal the underlying matter density fluctuations from the HI mapping, it is important to understand how HI gas traces the matter density distribution. Both nonlinear halo clustering and nonlinear effects modulating HI gas in halos may determine the scale below which the HI bias deviates from linearity. We employ three approaches to generate the mock HI density from a large-scale N-body simulation at low redshifts, and demonstrate that the assumption of HI linearity is valid at the scale corresponding to the first peak of baryon acoustic oscillations, but breaks down at $k \gtrsim 0.1\,h\, {\rm Mpc}^{-1}$. The nonlinear effects of halo clustering and HI content modulation counteract each other at small scales, and their competition results in a model-dependent "sweet-spot" redshift near $z$=1 where the HI bias is scale-independent down to small scales. We also find that the linear HI bias scales approximately linearly with redshift for $z\le 3$.

astro-ph.CO↗