Searcharxiv⌕ Search

arXiv subjects

Daeyoung Lee

Publications and source records attributed to Daeyoung Lee.

8 recordsLinked to original sources

RankShift: Evaluating Ranking Instability Across Benchmark Dimensions in Document Parsing

Document parsing benchmarks are often treated as stable measures of model quality, yet rankings may depend substantially on benchmark design choices. In this work, we present RankShift, a systematic study comparing OCR and vision-language models (VLMs) under controlled variation in dataset composition, matching strategy, and scoring methodology. Using OmniDocBench as a case study, we evaluate the ranking stability of 16 models across 3 dataset configurations, 6 matching strategies, and 3 scoring metrics. We quantify rank stability using pairwise rank correlation, top-k churn, and pairwise reversal rate. Our results show that the interaction between matching strategy and ground truth format is a substantial but understudied driver of instability: scoring against OmniDocBench's JSON annotations versus its Markdown rendering yields a mean Kendall's tau of approximately 0.33 under region-level matching strategies, with individual pairs as low as tau = 0.23, compared with tau = 0.77 when each page is scored as a single concatenated string. Dataset composition produces comparable instability across page-type strata (mean tau = 0.48) and layout-type strata (mean tau = 0.49), while the cross-dataset shift from OmniDocBench to Real5's Scanning reconstruction of the same pages is more stable (tau = 0.80). Metric choice is the most stable dimension, with mean tau of approximately 0.78 across NED, BLEU, and METEOR. These findings indicate that leaderboard position is not solely a property of model capability, motivating leaderboards that report not only headline scores but also the robustness of rankings across reasonable design choices.

cs.AI↗

VS-Flash: Harnessing Vector Search with Computational Memory-Semantic Flash

As generative AI continues to evolve, retrieval-augmented generation (RAG) systems are driving rapid growth in vector database capacity requirements. As datasets outgrow host DRAM, vector databases rely on MMap-based serving over SSDs to extend capacity. However, high-fan-out ANN searches generate concurrent irregular accesses that contend for host memory and storage resources, degrading bulk throughput and inflating tail latency under mixed workloads. In this paper, we address these challenges through memory-semantic capacity and workload-aware compute placement. VS-Flash exposes flash as a large-capacity memory-semantic space and dynamically selects the execution engine according to runtime query and workload characteristics. It offloads throughput-oriented large-batch queries to a Computational Memory-Semantic Flash (CMS-Flash) device, while executing latency-sensitive small-batch queries on the host CPU. Both paths access the same resident segments, avoiding a separate host-DRAM replica. Across three billion-scale datasets, VS-Flash improves bulk-search throughput over vanilla Milvus by 36.8X on average, with up to a 57.4X improvement. Under mixed workloads, VS-Flash reduces the p99 latency of latency-sensitive requests by up to 24.8X while sustaining device-side bulk search. These results show that VS-Flash delivers scalable vector search beyond host DRAM capacity while jointly improving bulk throughput and small-request tail latency.

cs.DB↗

Event Horizon Telescope Pattern Speeds in the Visibility Domain

The Event Horizon Telescope is preparing to produce time sequences of black hole images, or movies. In anticipation, we developed an autocorrelation technique to measure apparent rotational motion using the image-domain pattern speed $Ω_p$. Here, we extend this technique to the visibility domain and introduce the visibility amplitude pattern speed $Ω_{\mathrm{VA}}$. We show that in the Illinois v3 library of EHT source models, $Ω_{\mathrm{VA}}$ depends on the source inclination, black hole mass, black hole spin, accretion state (MAD or SANE), and baseline length, and then provide approximate fits for this dependence. We show that $Ω_{\mathrm{VA}}$ is particularly sensitive to baseline length for MAD (strongly magnetized) models, and that the slope of this dependence can be used to constrain black hole spin. As with $Ω_p$, models predict that $Ω_{\mathrm{VA}}$ is well below the Keplerian frequency in the emission region for all model parameters. This is consistent with the idea that $Ω_{\mathrm{VA}}$ measures an angular phase speed for waves propagating through the emission region. Finally, we identify the information that would be provided by space-based millimeter VLBI such as the proposed BHEX mission.

astro-ph.HE↗

Inferring Meteoroid Properties with Dynamic Nested Sampling: A Case Study of Orionid and Capricornid Shower Meteors

Accurate estimation of meteoroid bulk density is crucial for assessing spacecraft impact hazards from sub-millimeter to millimeter-sized meteoroids. Previous studies often used manual tuning or optimization methods to fit ablation and fragmentation models to optical meteor data, but subjective choices made physical properties and uncertainties difficult to compare. We develop a global, statistically robust method that uses Dynamic Nested Sampling to fit an erosion-fragmentation model to meteor light curves and deceleration measured by the Canadian Automated Meteor Observatory (CAMO) mirror tracking system and Electron-Multiplied CCD (EMCCD) cameras. Applied to 15 shower meteors, the method returns posterior distributions and Bayesian evidences for single- and double-fragmentation scenarios. Tests on four synthetic cases recover the known inputs, with best-guess solutions matching the true parameters. For 9 Orionids and 6 Alpha Capricornids with masses 1e-6 to 1e-5 kg, the median bulk density is 159 (+558/-57) kg/m3 for Orionids and 333 (+1089/-114) kg/m3 for Alpha Capricornids. Orionids are consistent with low-density cometary material, while Alpha Capricornids are systematically denser and show a second density cluster near 1300 kg/m3, consistent with higher-density asteroidal material. This framework enables automated, statistically rigorous characterization of meteoroid properties and will be extended to larger samples of shower and sporadic meteors across orbital classes.

astro-ph.EP↗

Rotation in Event Horizon Telescope Movies

The Event Horizon Telescope (EHT) has produced images of M87* and Sagittarius A*, and will soon produce time sequences of images, or movies. In anticipation of this, we describe a technique to measure the rotation rate, or pattern speed $Ω_p$, from movies using an autocorrelation technique. We validate the technique on Gaussian random field models with a known rotation rate and apply it to a library of synthetic images of Sgr A* based on general relativistic magnetohydrodynamics simulations. We predict that EHT movies will have $Ω_p \approx 1^\circ$ per $GMc^{-3}$, which is of order $15\%$ of the Keplerian orbital frequency in the emitting region. We can plausibly attribute the slow rotation seen in our models to the pattern speed of inward-propagating spiral shocks. We also find that $Ω_p$ depends strongly on inclination. Application of this technique will enable us to compare future EHT movies with the clockwise rotation of Sgr A* seen in near-infrared flares by GRAVITY. Pattern speed analysis of future EHT observations of M87* and Sgr A* may also provide novel constraints on black hole inclination and spin, as well as an independent measurement of black hole mass.

astro-ph.HE↗

Measures of Variance on Windowed Gaussian Processes

The variance and fractional variance on a fixed time window (variously known as "rms percent" or "modulation index") are commonly used to characterize the variability of astronomical sources. We summarize properties of this statistic for a Gaussian process.

astro-ph.IM↗

Millimeter light curves of Sagittarius A* observed during the 2017 Event Horizon Telescope campaign

The Event Horizon Telescope (EHT) observed the compact radio source, Sagittarius A* (Sgr A*), in the Galactic Center on 2017 April 5-11 in the 1.3 millimeter wavelength band. At the same time, interferometric array data from the Atacama Large Millimeter/submillimeter Array and the Submillimeter Array were collected, providing Sgr A* light curves simultaneous with the EHT observations. These data sets, complementing the EHT very-long-baseline interferometry, are characterized by a cadence and signal-to-noise ratio previously unattainable for Sgr A* at millimeter wavelengths, and they allow for the investigation of source variability on timescales as short as a minute. While most of the light curves correspond to a low variability state of Sgr A*, the April 11 observations follow an X-ray flare, and exhibit strongly enhanced variability. All of the light curves are consistent with a red noise process, with a power spectral density (PSD) slope measured to be between -2 and -3 on timescales between 1 min and several hours. Our results indicate a steepening of the PSD slope for timescales shorter than 0.3 h. The spectral energy distribution is flat at 220 GHz and there are no time-lags between the 213 and 229 GHz frequency bands, suggesting low optical depth for the event horizon scale source. We characterize Sgr A*'s variability, highlighting the different behavior observed just after the X-ray flare, and use Gaussian process modeling to extract a decorrelation timescale and a PSD slope. We also investigate the systematic calibration uncertainties by analyzing data from independent data reduction pipelines.

astro-ph.HE↗

Disks as Inhomogeneous, Anisotropic Gaussian Random Fields

We model astrophysical disk surface brightness fluctuations as an inhomogeneous, anisotropic, time-dependent Gaussian random field. The field locally obeys the stochastic partial differential equation of a Matérn field, which has a power spectrum that is flat at large scales and falls off as a power law at small scales. We provide a series of pedagogical examples and along the way provide a convenient parameterization for the local covariance. We then consider two applications to disks. In the first we generate a movie of a disk. In the second, by integrating over a movie of a disk, we generate synthetic light curves and show that the high frequency slope of the resulting power spectrum depends on the local covariance model. We finish with a summary and a brief discussion of other possible astrophysical applications.

astro-ph.IM↗