arXiv subjects

Benjamin Zhang

Publications and source records attributed to Benjamin Zhang.

10 recordsLinked to original sources

arXiv2026

Optimizing Irreversible Perturbations of the Unadjusted Langevin Algorithm

Irreversible perturbations accelerate the convergence of Langevin dynamics, breaking detailed balance while preserving the invariant measure. The design of optimal irreversible perturbations has been studied in the continuous-time Gaussian setting, but extensions to non-Gaussian target distributions, and the impact of time discretization on the design of optimal perturbations, have not been well understood. Numerical discretizations of Langevin dynamics introduce bias, which is typically exacerbated by irreversible perturbations; handling this interaction demands a joint treatment of acceleration and accuracy. This paper develops a systematic framework for optimizing position-independent irreversible perturbations of the unadjusted Langevin algorithm (ULA). We formulate a constrained optimization problem that simultaneously accounts for mixing efficiency and discretization bias, where the former is characterized by a spectral gap analogue and the latter is quantified via a weighted expected squared jump distance. Within this framework, we derive an explicit characterization of the optimal position-independent irreversible perturbation. Extensive numerical experiments demonstrate that our design yields faster convergence with controlled bias, and improves mean squared estimation errors compared to other choices of irreversible perturbation.

math.NA↗

arXiv2026

SVI-Bench: A Dynamic Microworld for Strategic Video Intelligence

True video intelligence demands more than recognizing what is visible: it requires reasoning about why events unfold, predicting what would change under different conditions, and deciding what to do next. We refer to this progression, from perception through causal reasoning and simulation to strategic planning, as Strategic Video Intelligence (SVI). No existing benchmark evaluates this capability stack: in-the-wild videos lack verifiable ground truth for causal and strategic questions, while synthetic environments sacrifice the complexity of real multi-agent systems. To bridge this gap, we introduce SVI-Bench, a large-scale benchmark that leverages team sports as a dynamic microworld, combining the complexity of real-world multi-agent interaction (10-22 agents making coordinated decisions under adversarial pressure) with the verifiability of explicit rules and definitive outcomes. SVI-Bench comprises approximately 35K hours of broadcast video, 15M annotated actions, 15K hours of expert commentary, 23K game reports, and 103K structured statistical records across basketball, soccer, and hockey, all constructed via a data engine that transforms raw game data into a dense, cross-referenced corpus. We organize evaluation into 9 tasks spanning a progressive four-pillar hierarchy: Dynamic Scene Understanding, Causal Reasoning, Strategic Simulation, and Agentic Synthesis. Evaluating strong multimodal and agentic baselines, we find a capability cliff: models perform competently on perceptual tasks, achieving approximately 74% on fine-grained action QA, but degrade sharply at each successive cognitive level. Agentic tasks prove hardest: the strongest model achieves only 5% accuracy when required to autonomously gather and integrate evidence across a corpus of 1.8M clips.

cs.CV↗

arXiv2025

Knothe-Rosenblatt maps via soft-constrained optimal transport

In the theory of optimal transport, the Knothe-Rosenblatt (KR) rearrangement provides an explicit construction to map between two probability measures by building one-dimensional transformations from the marginal conditionals of one measure to the other. The KR map has shown to be useful in different realms of mathematics and statistics, from proving functional inequalities to designing methodologies for sampling conditional distributions. It is known that the KR rearrangement can be obtained as the limit of a sequence of optimal transport maps with a weighted quadratic cost. We extend these results in this work by showing that one can obtain the KR map as a limit of maps that solve a relaxation of the weighted-cost optimal transport problem with a soft-constraint for the target distribution. In addition, we show that this procedure also applies to the construction of triangular velocity fields via dynamic optimal transport yielding optimal velocity fields. This justifies various variational methodologies for estimating KR maps in practice by minimizing a divergence between the target and pushforward measure through an approximate map. Moreover, it opens the possibilities for novel static and dynamic OT estimators for KR maps.

math.OC↗

arXiv2025

Cross-correlations between the CLAMATO Lyman-alpha forest and galaxies within the COSMOS field

We compute the 3D cross-correlation between the absorption of the $z\sim 2.3$ Lyman-alpha forest measured by the COSMOS Lyman-Alpha Mapping And Tomography Observations (CLAMATO) survey, and 1642 foreground galaxies with spectroscopic redshifts from several different surveys, including 3D-HST, CLAMATO, zCOSMOS-deep, MOSDEF, and VUDS. For each survey, we compare the measured cross-correlation with models incorporating the galaxy linear bias as well as observed redshift dispersion and systematic redshift offset. The derived redshift dispersion and offsets are generally consistent with those expected from, e.g., spectroscopic redshifts measured with UV absorption lines or NIR emission lines observed with specific instruments, but we find hints of `fingers-of-god' caused by overdensities in the field. We combine our foreground galaxy sample, and split them into 3 bins of robustly-estimated stellar mass in order to study the stellar mass-halo mass relationship. For sub-samples with median stellar masses of $\log_{10}(M_* / M_\odot) = [9.28,9.74,10.22]$, we find galaxy biases of $b_g\approx [2.1, 3.2,3.8]$, respectively. A comparison with mock measurements from the Bolshoi-Planck $N$-body simulation yields corresponding halo masses of $\log_{10}(M_* / M_\odot) \approx [10.5,11.7,12.1]$ for these stellar mass bins. At the low mass end, our results suggest enhanced star formation histories in mild tension with predictions from previous angular correlation and abundance matching-based observations, and the IllustrisTNG simulation.

astro-ph.CO↗

arXiv2025

A Fast Bayesian Method for Coherent Gravitational Wave Searches with Relative Astrometry

Using relative stellar astrometry for the detection of coherent gravitational wave sources is a promising method for the microhertz range, where no dedicated detectors currently exist. Compared to other gravitational wave detection techniques, astrometry operates in an extreme high-baseline-number and low-SNR-per-baseline limit, which leads to computational difficulties when using conventional Bayesian search techniques. We extend a technique for efficiently searching pulsar timing array datasets through the precomputation of inner products in the Bayesian likelihood, showing that it is applicable to astrometric datasets. Using this technique, we are able to reduce the total dataset size by up to a factor of $\mathcal{O}(100)$, while remaining accurate to within 1% over two orders of magnitude in gravitational wave frequency. Applying this technique to simulated astrometric datasets for the Kepler Space Telescope and Nancy Grace Roman Space Telescope missions, we obtain forecasts for the sensitivity of these missions to coherent gravitational waves. Due to the low angular sky coverage of astrometric baselines, we find that coherent gravitational wave sources are poorly localized on the sky. Despite this, from $10^{-8}$ Hz to $10^{-6}$ Hz, we find that Roman is sensitive to coherent gravitational waves with an instantaneous strain above $h_0 \simeq 10^{-11.4}$, and Kepler is sensitive to strains above $h_0 \simeq $ $10^{-12.4}$. At this strain, we can detect a source with a frequency of $10^{-7}$ Hz and a chirp mass of $10^9$ $M_\odot$ at a luminosity distance of 3.6 Mpc for Kepler, and 0.3 Mpc for Roman.

astro-ph.IM↗

arXiv2025

ExAct: A Video-Language Benchmark for Expert Action Analysis

We present ExAct, a new video-language benchmark for expert-level understanding of skilled physical human activities. Our new benchmark contains 3521 expert-curated video question-answer pairs spanning 11 physical activities in 6 domains: Sports, Bike Repair, Cooking, Health, Music, and Dance. ExAct requires the correct answer to be selected from five carefully designed candidate options, thus necessitating a nuanced, fine-grained, expert-level understanding of physical human skills. Evaluating the recent state-of-the-art VLMs on ExAct reveals a substantial performance gap relative to human expert performance. Specifically, the best-performing GPT-4o model achieves only 44.70% accuracy, well below the 82.02% attained by trained human specialists/experts. We believe that ExAct will be beneficial for developing and evaluating VLMs capable of precise understanding of human skills in various physical and procedural domains. Dataset and code are available at https://texaser.github.io/exact_project_page/

cs.CV↗

arXiv2024

Zooming by in the CARPoolGP lane: new CAMELS-TNG simulations of zoomed-in massive halos

Galaxy formation models within cosmological hydrodynamical simulations contain numerous parameters with non-trivial influences over the resulting properties of simulated cosmic structures and galaxy populations. It is computationally challenging to sample these high dimensional parameter spaces with simulations, particularly for halos in the high-mass end of the mass function. In this work, we develop a novel sampling and reduced variance regression method, CARPoolGP, which leverages built-in correlations between samples in different locations of high dimensional parameter spaces to provide an efficient way to explore parameter space and generate low variance emulations of summary statistics. We use this method to extend the Cosmology and Astrophysics with MachinE Learning Simulations (CAMELS) to include a set of 768 zoom-in simulations of halos in the mass range of $10^{13} - 10^{14.5} M_\odot\,h^{-1}$ that span a 28-dimensional parameter space in the IllustrisTNG model. With these simulations and the CARPoolGP emulation method, we explore parameter trends in the Compton $Y-M$, black hole mass-halo mass, and metallicity-mass relations, as well as thermodynamic profiles and quenched fractions of satellite galaxies. We use these emulations to provide a physical picture of the complex interplay between supernova and active galactic nuclei feedback. We then use emulations of the $Y-M$ relation of massive halos to perform Fisher forecasts on astrophysical parameters for future Sunyaev-Zeldovich observations and find a significant improvement in forecasted constraints. We publicly release both the simulation suite and CARPoolGP software package.

astro-ph.GA↗

arXiv2022

Alignments Between Galaxies and the Cosmic Web at $z \sim 1-2$ in the IllustrisTNG Simulations

Galaxy formation theories predict that galaxy shapes and angular momenta have non-random alignments with the cosmic web. This leads to so-called intrinsic alignment between pairs of galaxies, which is important to quantify as a nuisance parameter for weak lensing. We study galaxy-cosmic web alignment in the IllustrisTNG suite of hydrodynamical simulations at redshifts 1 and 2, finding that alignment trends are consistent with previous studies. However, we find that the magnitude of the spin alignment signal is $\sim 2.4 \times$ weaker than seen in previous studies of the Horizon-AGN simulation, suggesting that this signal may have significant dependence on subgrid physics. Based on IllustrisTNG, we then construct mock observational spectroscopic surveys that can probe shape-cosmic web alignment at $z \sim 1-2$, modeled on the low-$z$ galaxy redshift and IGM tomography surveys on the upcoming Subaru Prime Focus Spectrograph Galaxy Evolution (PFS GE) survey. However, even over box sizes of $L=205\,h^{-1}\,\mathrm{Mpc}$, we find that global anisotropies induce a sample variance in the 2D projected alignment signal that depend on the projected direction -- this induces significant errors in the observed alignment. We predict a $5.3\sigma$ detection of IllustrisTNG's shape alignment signal at $z \sim 1$ from Subaru PFS GE, although a detection would be challenging at $z \sim 2$. However, a rough rescaling of the relative alignment signal strengths between the TNG and HorizonAGN simulations suggests that PFS GE should be able to more easily constrain the latter's stronger signal.

astro-ph.CO↗

arXiv2021

A Koopman framework for rare event simulation in stochastic differential equations

We exploit the relationship between the stochastic Koopman operator and the Kolmogorov backward equation to construct importance sampling schemes for stochastic differential equations. Specifically, we propose using eigenfunctions of the stochastic Koopman operator to approximate the Doob transform for an observable of interest (e.g., associated with a rare event) which in turn yields an approximation of the corresponding zero-variance importance sampling estimator. Our approach is broadly applicable and systematic, treating non-normal systems, non-gradient systems, and systems with oscillatory dynamics or rank-deficient noise in a common framework. In nonlinear settings where the stochastic Koopman eigenfunctions cannot be derived analytically, we use dynamic mode decomposition (DMD) methods to compute them numerically, but the framework is agnostic to the particular numerical method employed. Numerical experiments demonstrate that even coarse approximations of a few eigenfunctions, where the latter are built from non-rare trajectories, can produce effective importance sampling schemes for rare events.

stat.CO↗

arXiv2020

TARDIS Paper II: Synergistic Density Reconstruction from Lyman-alpha Forest and Spectroscopic Galaxy Surveys with Applications to Protoclusters and the Cosmic Web

In this work we expand upon the Tomographic Absorption Reconstruction and Density Inference Scheme (TARDIS) in order to include multiple tracers while reconstructing matter density fields at Cosmic Noon (z ~ 2-3). In particular, we jointly reconstruct the underlying density field from simulated Lyman-Alpha forest observations at $z\sim 2.5$ and an overlapping galaxy survey. We find that these data are synergistic, with the Lyman Alpha forest providing reconstruction of low density regions and galaxy surveys tracing the density peaks. We find a more accurate power spectra reconstruction going to higher scales when fitting these two data-sets simultaneously than if using either one individually. When applied to cosmic web analysis, we find performing the joint analysis is equivalent to a Lyman Alpha survey with significantly increased sight-line spacing. Since we reconstruct the velocity field and matter field jointly, we demonstrate the ability to evolve the mock observed volume further to z=0, allowing us to create a rigorous definition of "proto-cluster" as regions which will evolve into clusters. We apply our reconstructions to study protocluster structure and evolution, finding for realistic survey parameters we can provide accurate mass estimates of the z \approx 2 structures and their z = 0 fate.

astro-ph.CO↗