arXiv subjects

Yunlin Zeng

Publications and source records attributed to Yunlin Zeng.

10 recordsLinked to original sources

arXiv2026

Bayesian Joint Velocity and Impedance Inversion via Diffusion Models Conditioned on Common Image Gathers

We present a multi-parameter simulation-based inference framework for joint Bayesian recovery of subsurface velocity and acoustic impedance from seismic data. A score-based diffusion model is conditioned on two complementary Common Image Gathers (CIGs): an inverse-scattering CIG encoding reflectivity amplitude and an anti-ISIC CIG encoding kinematic velocity errors. The model simultaneously samples the posterior distributions of both parameters. Training labels are deliberately decoupled to prevent the model from exploiting the Gardner relationship: velocity targets are lightly smoothed to match the long-wavelength content of the anti-ISIC CIG, while impedance targets retain the unsmoothed ground truth. On the Compass benchmark the model achieves velocity SSIM of 0.967 (RMSE 0.050 km/s) and impedance SSIM 0.867 (RMSE 0.279 km/s g/cm^3), with velocity quality confirmed by CIG focusing.

physics.geo-ph↗

arXiv2026

SPoRC-VIST: A Benchmark for Evaluating Generative Natural Narrative in Vision-Language Models

Vision-Language Models (VLMs) have achieved remarkable success in descriptive tasks such as image captioning and visual question answering (VQA). However, their ability to generate engaging, long-form narratives -- specifically multi-speaker podcast dialogues -- remains under-explored and difficult to evaluate. Standard metrics like BLEU and ROUGE fail to capture the nuances of conversational naturalness, personality, and narrative flow, often rewarding safe, repetitive outputs over engaging storytelling. In this work, we present a novel pipeline for end-to-end visual podcast generation, and fine-tune a Qwen3-VL-32B model on a curated dataset of 4,000 image-dialogue pairs. Crucially, we use a synthetic-to-real training strategy: we train on high-quality podcast dialogues from the Structured Podcast Research Corpus (SPoRC) paired with synthetically generated imagery, and evaluate on real-world photo sequences from the Visual Storytelling Dataset (VIST). This rigorous setup tests the model's ability to generalize from synthetic training data to real-world visual domains. We propose a comprehensive evaluation framework that moves beyond textual overlap, and use AI-as-a-judge (Gemini 3 Pro, Claude Opus 4.5, GPT 5.2) and novel style metrics (average turn length, speaker switch rate) to assess quality. Our experiments demonstrate that our fine-tuned 32B model significantly outperforms a 235B base model in conversational naturalness ($>$80\% win rate) and narrative depth (+50\% turn length), while maintaining identical visual grounding capabilities (CLIPScore: 20.39).

cs.LG↗

arXiv2025

Power-scaled Bayesian Inference with Score-based Generative Models

We propose a score-based generative algorithm for sampling from power-scaled priors and likelihoods within the Bayesian inference framework. Our algorithm enables flexible control over prior-likelihood influence without requiring retraining for different power-scaling configurations. Specifically, we focus on synthesizing seismic velocity models conditioned on imaged seismic. Our method enables sensitivity analysis by sampling from intermediate power posteriors, allowing us to assess the relative influence of the prior and likelihood on samples of the posterior distribution. Through a comprehensive set of experiments, we evaluate the effects of varying the power parameter in different settings: applying it solely to the prior, to the likelihood of a Bayesian formulation, and to both simultaneously. The results show that increasing the power of the likelihood up to a certain threshold improves the fidelity of posterior samples to the conditioning data (e.g., seismic images), while decreasing the prior power promotes greater structural diversity among samples. Moreover, we find that moderate scaling of the likelihood leads to a reduced shot data residual, confirming its utility in posterior refinement.

cs.LG↗

arXiv2025

Full-waveform variational inference with full common-image gathers and diffusion network

Accurate seismic imaging and velocity estimation are essential for subsurface characterization. Conventional inversion techniques, such as full-waveform inversion, remain computationally expensive and sensitive to initial velocity models. To address these challenges, we propose a simulation-based inference framework with conditional elucidated diffusion models for posterior velocity-model sampling. Our approach incorporates both horizontal and vertical subsurface offset common-image gathers to capture a broader range of reflector geometries, including gently dipping structures and steep dipping layers. Additionally, we introduce the background-velocity model as an input condition to enhance generalization across varying geological settings. We evaluate our method on the SEAM dataset, which features complex salt geometries, using a patch-based training approach. Experimental results demonstrate that adding the background-velocity model as an additional conditioning variable significantly enhances performance, improving SSIM from $0.717$ to $0.733$ and reducing RMSE from $0.381\,$km/s to $0.274\,$km/s. Furthermore, uncertainty quantification analysis shows that our proposed approach yields better-calibrated uncertainty estimates, reducing uncertainty calibration error from $6.68\,$km/s to $3.91\,$km/s. These results show robust amortized seismic inversion with uncertainty quantification.

physics.geo-ph↗

arXiv2024

Machine learning-enabled velocity model building with uncertainty quantification

Accurately characterizing migration velocity models is crucial for a wide range of geophysical applications, from hydrocarbon exploration to monitoring of CO2 sequestration projects. Traditional velocity model building methods such as Full-Waveform Inversion (FWI) are powerful but often struggle with the inherent complexities of the inverse problem, including noise, limited bandwidth, receiver aperture and computational constraints. To address these challenges, we propose a scalable methodology that integrates generative modeling, in the form of Diffusion networks, with physics-informed summary statistics, making it suitable for complicated imaging problems including field datasets. By defining these summary statistics in terms of subsurface-offset image volumes for poor initial velocity models, our approach allows for computationally efficient generation of Bayesian posterior samples for migration velocity models that offer a useful assessment of uncertainty. To validate our approach, we introduce a battery of tests that measure the quality of the inferred velocity models, as well as the quality of the inferred uncertainties. With modern synthetic datasets, we reconfirm gains from using subsurface-image gathers as the conditioning observable. For complex velocity model building involving salt, we propose a new iterative workflow that refines amortized posterior approximations with salt flooding and demonstrate how the uncertainty in the velocity model can be propagated to the final product reverse time migrated images. Finally, we present a proof of concept on field datasets to show that our method can scale to industry-sized problems.

cs.LG↗

arXiv2021

The Gliese 86 Binary System: A Warm Jupiter Formed in a Disk Truncated at $\approx$2 AU

Gliese 86 is a nearby K dwarf hosting a giant planet on a $\approx$16-day orbit and an outer white dwarf companion on a $\approx$century-long orbit. In this study we combine radial velocity data (including new measurements spanning more than a decade) with high angular resolution imaging and absolute astrometry from Hipparcos and Gaia to measure the current orbits and masses of both companions. We then simulate the evolution of the Gl 86 system to constrain its primordial orbit when both stars were on the main sequence; the closest approach between the two stars was then about $9\,$AU. Such a close separation limited the size of the protoplanetary disk of Gl 86 A and dynamically hindered the formation of the giant planet around it. Our measurements of Gl 86 B and Gl 86 Ab's orbits reveal Gl 86 as a system in which giant planet formation took place in a disk truncated at $\approx$2$\,$AU. Such a disk would be just big enough to harbor the dust mass and total mass needed to assemble Gl 86 Ab's core and envelope, assuming a high disk accretion rate and a low viscosity. Inefficient accretion of the disk onto Gl 86 Ab, however, would require a disk massive enough to approach the Toomre stability limit at its outer truncation radius. The orbital architecture of the Gl 86 system shows that giant planets can form even in severely truncated disks and provides an important benchmark for planet formation theory.

astro-ph.EP↗

arXiv2021

Precise Masses and Orbits for Nine Radial Velocity Exoplanets

Radial velocity (RV) surveys have discovered hundreds of exoplanetary systems but suffer from a fundamental degeneracy between planet mass $M_p$ and orbital inclination $i$. In this paper we break this degeneracy by combining RVs with complementary absolute astrometry taken from the Gaia EDR3 version of the cross-calibrated Hipparcos-Gaia Catalog of Accelerations (HGCA). We use the Markov Chain Monte Carlo orbit code $\tt orvara$ to simultaneously fit literature RVs and absolute astrometry from the HGCA. We constrain the orbits, masses, and inclinations of nine single and massive RV companions orbiting nearby G and K stars. We confirm the planetary nature of six companions: HD 29021 b ($4.47_{-0.65}^{+0.67}\,M_{\rm Jup}$), HD 81040 b ($7.24_{-0.37}^{+1.0}\,M_{\rm Jup}$), HD 87883 b ($6.31_{-0.32}^{+0.31}\,M_{\rm Jup}$), HD 98649 b ($9.7_{-1.9}^{+2.3}\,M_{\rm Jup}$), HD 106252 b ($10.00_{-0.73}^{+0.78}\,M_{\rm Jup}$), and HD 171238 b ($8.8_{-1.3}^{+3.6}\,M_{\rm Jup}$). We place one companion, HD 196067 b ($12.5_{-1.8}^{+2.5}\,M_{\rm Jup}$) on the planet-brown dwarf boundary, and two companions in the low mass brown dwarf regime: HD 106515 Ab ($18.9_{-1.4}^{+1.5}\,M_{\rm Jup}$), and HD 221420 b (${20.6}_{-1.6}^{+2.0}\,M_{\rm Jup}$). The brown dwarf HD 221420 b, with a semi-major axis of ${9.99}_{-0.70}^{+0.74}$ AU, a period of ${27.7}_{-2.5}^{+3.0}$ years, and an eccentricity of $0.162_{-0.030}^{+0.035}$ represents a promising target for high-contrast imaging. The RV orbits of HD 87883 b, HD 98649 b, HD 171238 b, and HD 196067 b are not fully constrained yet because of insufficient RV data. We find two possible inclinations for each of these orbits due to difficulty in separating prograde from retrograde orbits, but we expect this will change decisively with future Gaia data releases.

astro-ph.EP↗

arXiv2021

htof: A new open-source tool for analyzing Hipparcos, Gaia, and future astrometric missions

We present htof, an open-source tool for interpreting and fitting the intermediate astrometric data (IAD) from both the 1997 and 2007 reductions of Hipparcos, the scanning-law of Gaia, and future missions such as the Nancy Grace Roman Space Telescope (NGRST). htof solves for the astrometric parameters of any system for any arbitrary combination of absolute astrometric missions. In preparation for later Gaia data releases, htof supports arbitrarily high-order astrometric solutions (e.g. five-, seven-, nine-parameter fits). Using htof, we find that the IAD of 6617 sources in Hipparcos 2007 might have been affected by a data corruption issue. htof integrates an ad-hoc correction that reconciles the IAD of these sources with their published catalog solutions. We developed htof to study masses and orbital parameters of sub-stellar companions, and we outline its implementation in one orbit fitting code (orvara, https://github.com/t-brandt/orvara). We use htof to predict a range of hypothetical additional planets in the $β$~Pic system, which could be detected by coupling NGRST astrometry with Gaia and Hipparcos. htof is pip installable and available at https://github.com/gmbrandt/htof .

astro-ph.IM↗

arXiv2021

orvara: An Efficient Code to Fit Orbits using Radial Velocity, Absolute, and/or Relative Astrometry

We present an open-source Python package, Orbits from Radial Velocity, Absolute, and/or Relative Astrometry (orvara), to fit Keplerian orbits to any combination of radial velocity, relative astrometry, and absolute astrometry data from the Hipparcos-Gaia Catalog of Accelerations. By combining these three data types, one can measure precise masses and sometimes orbital parameters even when the observations cover a small fraction of an orbit. orvara achieves its computational performance with an eccentric anomaly solver five to ten times faster than commonly used approaches, low-level memory management to avoid python overheads, and by analytically marginalizing out parallax, barycenter proper motion, and the instrument-specific radial velocity zero points. Through its integration with the Hipparcos and Gaia intermediate astrometry package htof, orvara can properly account for the epoch astrometry measurements of Hipparcos and the measurement times and scan angles of individual Gaia epochs. We configure orvara with modifiable .ini configuration files tailored to any specific stellar or planetary system. We demonstrate orvara with a case study application to a recently discovered white dwarf/main sequence (WD/MS) system, HD 159062. By adding absolute astrometry to literature RV and relative astrometry data, our comprehensive MCMC analysis improves the precision of HD 159062B's mass by more than an order of magnitude to $0.6083^{+0.0083}_{-0.0073}\,M_\odot$. We also derive a low eccentricity and large semimajor axis, establishing HD 159062AB as a system that did not experience Roche lobe overflow.

astro-ph.IM↗

arXiv2020

SCExAO/CHARIS Direct Imaging Discovery of a 20 au Separation, Low-Mass Ratio Brown Dwarf Companion to an Accelerating Sun-like Star

We present the direct imaging discovery of a substellar companion to the nearby Sun-like star, HD 33632 Aa, at a projected separation of $\sim$ 20 au, obtained with SCExAO/CHARIS integral field spectroscopy complemented by Keck/NIRC2 thermal infrared imaging. The companion, HD 33632 Ab, induces a 10.5$σ$ astrometric acceleration on the star as detected with the $Gaia$ and $Hipparcos$ satellites. SCExAO/CHARIS $JHK$ (1.1--2.4 $μ$m) spectra and Keck/NIRC2 $L_{\rm p}$ (3.78 $μ$m) photometry are best matched by a field L/T transition object: an older, higher gravity, and less dusty counterpart to HR 8799 cde. Combining our astrometry with $Gaia/Hipparcos$ data and archival Lick Observatory radial-velocities, we measure a dynamical mass of 46.4 $\pm$ 8 $M_{\rm J}$ and an eccentricity of $e$ $<$0.46 at 95\% confidence. HD 33632 Ab's mass and mass ratio (4.0\% $\pm$ 0.7\%) are comparable to the low-mass brown dwarf GJ 758 B and intermediate between the more massive brown dwarf HD 19467 B and the (near-)planet mass companions to HR 2562 and GJ 504. Using $Gaia$ to select for direct imaging observations with the newest extreme adaptive optics systems can reveal substellar or even planet-mass companions on solar system-like scales at an increased frequency compared to blind surveys.

astro-ph.SR↗