arXiv subjects

Jieun Choi

Publications and source records attributed to Jieun Choi.

At least 19 recordsLinked to original sources

arXiv2026

Programmable spatial coherence tomography: diffraction-limited three-dimensional reflection imaging under modulated monochromatic illumination

Depth sectioning in reflection microscopy has predominantly relied on temporal coherence gating. Here we show that volumetric reflection tomography at diffraction-limited resolution can be achieved under monochromatic illumination by engineering spatial, rather than temporal, coherence. In programmable spatial coherence tomography (PSCT), a sequence of pupil-coded illumination patterns with angular-spectrum diversity generates measurement redundancy enabling the system to calibrate itself, jointly retrieving aberrations, illumination profiles, and sample motion without guide stars or modal priors. We demonstrate label-free volumetric imaging of thick human tissues, organoids, frequency-resolved dynamic contrast, and high-resolution in vivo brain imaging through a cranial window. These results position PSCT as an alternative to temporal coherence based reflection imaging in complex biological systems.

physics.optics↗

arXiv2025

FineServe: Precision-Aware KV Slab and Two-Level Scheduling for Heterogeneous Precision LLM Serving

Recent advances in Post-Training Quantization (PTQ) techniques have significantly increased demand for serving quantized large language models (LLMs), enabling higher throughput and substantially reduced memory usage with minimal accuracy loss. Quantized models address memory constraints in LLMs and enhance GPU resource utilization through efficient GPU sharing. However, quantized models have smaller KV block sizes than non-quantized models, causing limited memory efficiency due to memory fragmentation. Also, distinct resource usage patterns between quantized and non-quantized models require efficient scheduling to maximize throughput. To address these challenges, we propose FineServe, an inference serving framework for mixed-precision LLMs. FineServe's key contributions include: (1) KV Slab, a precision-aware adaptive memory management technique dynamically allocating KV cache based on model quantization characteristics, significantly reducing GPU memory fragmentation, and (2) a two-level scheduling framework comprising a global scheduler that places models to GPUs based on request rates, latency SLOs, and memory constraints and efficiency, and a local scheduler that adaptively adjusts batch sizes according to real-time request fluctuations. Experimental results demonstrate that FineServe achieves up to 2.2x higher SLO attainment and 1.8x higher token generation throughput compared to the state-of-the-art GPU sharing systems.

cs.DC↗

arXiv2025

FireQ: Fast INT4-FP8 Kernel and RoPE-aware Quantization for LLM Inference Acceleration

As large language models become increasingly prevalent, memory bandwidth constraints significantly limit inference throughput, motivating post-training quantization (PTQ). In this paper, we propose FireQ, a co-designed PTQ framework and an INT4-FP8 matrix multiplication kernel that accelerates LLM inference across all linear layers. Specifically, FireQ quantizes linear layer weights and key-values to INT4, and activations and queries to FP8, significantly enhancing throughput. Additionally, we introduce a three-stage pipelining for the prefill phase, which modifies the FlashAttention-3 kernel, effectively reducing time-to-first-token in the prefill phase. To minimize accuracy loss from quantization, we develop novel outlier smoothing techniques tailored separately for linear and attention layers. In linear layers, we explicitly use per-tensor scaling to prevent underflow caused by the FP8 quantization scaling factor of INT4 quantization, and channel-wise scaling to compensate for coarse granularity of INT4. In attention layers, we address quantization challenges posed by rotary positional embeddings (RoPE) by combining pre-RoPE and post-RoPE scaling strategies. FireQ significantly outperforms state-of-the-art methods, achieving 1.68x faster inference in feed-forward network layers on Llama2-7B and 1.26x faster prefill phase performance on Llama3-8B compared to QServe, with negligible accuracy loss.

cs.LG↗

arXiv2025

Review learning: Real world validation of privacy preserving continual learning across medical institutions

When a deep learning model is trained sequentially on different datasets, it often forgets the knowledge learned from previous data, a problem known as catastrophic forgetting. This damages the model's performance on diverse datasets, which is critical in privacy-preserving deep learning (PPDL) applications based on transfer learning (TL). To overcome this, we introduce "review learning" (RevL), a low cost continual learning algorithm for diagnosis prediction using electronic health records (EHR) within a PPDL framework. RevL generates data samples from the model which are used to review knowledge from previous datasets. Six simulated institutional experiments and one real-world experiment involving three medical institutions were conducted to validate RevL, using three binary classification EHR data. In the real-world experiment with data from 106,508 patients, the mean global area under the receiver operating curve was 0.710 for RevL and 0.655 for TL. These results demonstrate RevL's ability to retain previously learned knowledge and its effectiveness in real-world PPDL scenarios. Our work establishes a realistic pipeline for PPDL research based on model transfers across institutions and highlights the practicality of continual learning in real-world medical settings using private EHR data.

cs.AI↗

arXiv2025

MGHanD: Multi-modal Guidance for authentic Hand Diffusion

Diffusion-based methods have achieved significant successes in T2I generation, providing realistic images from text prompts. Despite their capabilities, these models face persistent challenges in generating realistic human hands, often producing images with incorrect finger counts and structurally deformed hands. MGHanD addresses this challenge by applying multi-modal guidance during the inference process. For visual guidance, we employ a discriminator trained on a dataset comprising paired real and generated images with captions, derived from various hand-in-the-wild datasets. We also employ textual guidance with LoRA adapter, which learns the direction from `hands' towards more detailed prompts such as `natural hands', and `anatomically correct fingers' at the latent level. A cumulative hand mask which is gradually enlarged in the assigned time step is applied to the added guidance, allowing the hand to be refined while maintaining the rich generative capabilities of the pre-trained model. In the experiments, our method achieves superior hand generation qualities, without any specific conditions or priors. We carry out both quantitative and qualitative evaluations, along with user studies, to showcase the benefits of our approach in producing high-quality hand images.

cs.CV↗

arXiv2024

A Survey on Integration of Large Language Models with Intelligent Robots

In recent years, the integration of large language models (LLMs) has revolutionized the field of robotics, enabling robots to communicate, understand, and reason with human-like proficiency. This paper explores the multifaceted impact of LLMs on robotics, addressing key challenges and opportunities for leveraging these models across various domains. By categorizing and analyzing LLM applications within core robotics elements -- communication, perception, planning, and control -- we aim to provide actionable insights for researchers seeking to integrate LLMs into their robotic systems. Our investigation focuses on LLMs developed post-GPT-3.5, primarily in text-based modalities while also considering multimodal approaches for perception and control. We offer comprehensive guidelines and examples for prompt engineering, facilitating beginners' access to LLM-based robotics solutions. Through tutorial-level examples and structured prompt construction, we illustrate how LLM-guided enhancements can be seamlessly integrated into robotics applications. This survey serves as a roadmap for researchers navigating the evolving landscape of LLM-driven robotics, offering a comprehensive overview and practical guidance for harnessing the power of language models in robotics development.

cs.RO↗

arXiv2021

Learning to increase matching efficiency in identifying additional b-jets in the $\text{t}\bar{\text{t}}\text{b}\bar{\text{b}}$ process

The $\text{t}\bar{\text{t}}\text{H}(\text{b}\bar{\text{b}})$ process is an essential channel to reveal the Higgs properties but has an irreducible background from the $\text{t}\bar{\text{t}}\text{b}\bar{\text{b}}$ process, which produces a top quark pair in association with a b quark pair. Therefore, understanding the $\text{t}\bar{\text{t}}\text{b}\bar{\text{b}}$ process is crucial for improving the sensitivity of a search for the $\text{t}\bar{\text{t}}\text{H}(\text{b}\bar{\text{b}})$ process. To this end, when measuring the differential cross-section of the $\text{t}\bar{\text{t}}\text{b}\bar{\text{b}}$ process, we need to distinguish the b-jets originated from top quark decays, and additional b-jets originated from gluon splitting. Since there are no simple identification rules, we adopt deep learning methods to learn from data to identify the additional b-jets from the $\text{t}\bar{\text{t}}\text{b}\bar{\text{b}}$ events. Specifically, by exploiting the special structure of the $\text{t}\bar{\text{t}}\text{b}\bar{\text{b}}$ event data, we propose several loss functions that can be minimized to directly increase the matching efficiency, the accuracy of identifying additional b-jets. We discuss the difference between our method and another deep learning-based approach based on binary classification arXiv:1910.14535 using synthetic data. We then verify that additional b-jets can be identified more accurately by increasing matching efficiency directly rather than the binary classification accuracy, using simulated $\text{t}\bar{\text{t}}\text{b}\bar{\text{b}}$ event data in the lepton+jets channel from pp collision at $\sqrt{s}$ = 13 TeV.

hep-ex↗

arXiv2020

Identification of additional jets in the t-tbar b-bbar events using a deep neural network

In the top quark pair production in association with the Higgs boson decaying to a b quark pair t-tbar H (b-bbar), the final state has an irreducible nonresonant background from the production of a top quark pair in association with a b quark pair t-tbar b-bbar. Therefore, understanding of the t-tbar b-bbar process precisely in particular differential cross-section as functions of the properties of the additional b jets not from the top quark decay is essential for improving the sensitivity of a search for the t-tbar H b-bbar process. The two additional b jets can be identified by using various approaches. In this paper, the performances are compared quantitatively in the lepton+jets decay channel in terms of the matching efficiency of assigning two additional b jets as a figure of merit. This study provides valuable information towards the precise measurement of differential cross-sections as a function of properties of the additional b jets in the t-tbar b-bbar events. We showed that a matching efficiency of around 40% could be achieved using a deep neural network method. In the events with at least 4 b jets, this performance is 8% better than that achieved using minimum Delta R(b,bbar) method. This is consistent with the boosted decision tree method within its statistical uncertainty.

hep-ex↗

arXiv2020

Varying Electronic Coupling at Graphene-Copper Interfaces Probed with Raman Spectroscopy

As the synthesis of graphene on copper became one of the primary preparation methods for both fundamental research and industrial application, Raman spectra of graphene/Cu systems need to be quantitatively understood regarding how their interactions affect the electronic structure of graphene. Using multi-wavelength Raman spectroscopy, we investigated three types of graphene bound on Cu: graphene grown on Cu foils and Cu film/SiO2, and Cu-evaporated exfoliated graphene. 2D peak frequencies of the first two samples were ~17 cm-1 higher than expected for 1.96 eV excitation even when the effect of strain was considered. More notably, the upshift in 2D decreased with increasing excitation energy. Based on control experiments using Cu-evaporated graphene, we revealed that the spectral anomaly was induced by environment-dependent nonlinear dispersion in the electronic bands of graphene and determined the degree of the electronic modification. We also showed that the large upshifts of G and 2D peaks originating from differential thermal expansion of Cu could be significantly reduced by backing Cu films with dielectric substrates of insignificant thermal expansion. The quantitative analysis of electronic coupling between graphene and Cu presented in this study will be highly useful in characterizing as-grown graphene and possibly in other forms.

cond-mat.mtrl-sci↗

arXiv2019

Self-consistent predictions for LIER-like emission lines from post-AGB stars

Early type galaxies (ETGs) frequently show emission from warm ionized gas. These Low Ionization Emission Regions (LIERs) were originally attributed to a central, low-luminosity active galactic nuclei. However, the recent discovery of spatially-extended LIER emission suggests ionization by both a central source and an extended component that follows a stellar-like radial distribution. For passively-evolving galaxies with old stellar populations, hot post-Asymptotic Giant Branch (AGB) stars are the only viable extended source of ionizing photons. In this work, we present the first prediction of LIER-like emission from post-AGB stars that is based on fully self-consistent stellar evolution and photoionization models. We show that models where post-AGB stars are the dominant source of ionizing photons reproduce the nebular emission signatures observed in ETGs, including LIER-like emission line ratios in standard optical diagnostic diagrams and H$α$ equivalent widths of order 0.1-3 angstroms. We test the sensitivity of LIER-like emission to the details of post-AGB models, including the mass loss efficiency and convective mixing efficiency, and show that line strengths are relatively insensitive to post-AGB timescale variations. Finally, we examine the UV-optical colors of the models and the stellar populations responsible for the UV-excess observed in some ETGs. We find that allowing as little as 3% of the HB population to be uniformly distributed to very hot temperatures (30,000 K) produces realistic UV colors for old, quiescent ETGs.

astro-ph.GA↗

arXiv2019

The Imprint of Element Abundance Patterns on Quiescent Galaxy SEDs

Stellar population synthesis (SPS) models have long struggled to reproduce observed optical through near-IR (NIR) spectral energy distributions (SED) of massive quiescent galaxies. We revisit this issue using a novel approach that combines the diagnostic power of full-spectrum fitting with recently updated stellar spectral libraries. First, we perform full-spectrum fitting of continuum-normalized stacked SDSS spectra in bins of velocity dispersion to infer their stellar population properties, such as the elemental abundances and age. Next, we use the resulting best-fit parameters to compute $ugriz$ colors, which are then compared to observed colors of the same galaxies. With this approach we are able to predict the $ugriz$ SEDs of low and high mass galaxies at the $\lesssim 0.03$ mag level in nearly all cases. We find that the full optical through NIR SEDs of quiescent galaxies can be reproduced only when the spectrum is fit with a flexibility that is able to capture the behavior of the entire optical absorption line spectrum. The models include variations in individual elemental abundances, nebular emission lines, and the presence of young stellar components. The successful prediction of the SED shape from continuum-normalized spectra implies that the continuum information is largely contained in the narrow absorption features. These results also imply that attempts to model broadband photometry of quiescent systems will suffer from potentially significant biases if the detailed abundance patterns are not taken into account.

astro-ph.GA↗

arXiv2019

A Novel Approach to Constrain Rotational Mixing & Convective-Core Overshoot in Stars Using the Initial-Final Mass Relation

The semi-empirical initial-final mass relation (IFMR) connects spectroscopically analyzed white dwarfs in star clusters to the initial masses of the stars that formed them. Most current stellar evolution models, however, predict that stars will evolve to white dwarfs $\sim$0.1 M$_\odot$ less massive than that found in the IFMR. We first look at how varying theoretical mass-loss rates, third dredge-up efficiencies, and convective-core overshoot may help explain the differences between models and observations. These parameters play an important role at the lowest masses (M$_{\rm initial}$ $<$ 3 M$_\odot$). At higher masses, only convective-core overshoot meaningfully affects white dwarf mass, but alone it likely cannot explain the observed white dwarf masses nor why the IFMR scatter is larger than observational errors predict. These higher masses, however, are also where rotational mixing in main sequence stars begins to create more massive cores, and hence more massive white dwarfs. This rotational mixing also extends a star's lifetime, making faster rotating progenitors appear like less massive stars in their semi-empirical age analysis. Applying the observed range of young B-dwarf rotations to the MIST or SYCLIST rotational models demonstrates a marked improvement in reproducing both the observed IFMR data and its scatter. The incorporation of both rotation and efficient convective-core overshoot significantly improves the match with observations. This work shows that the IFMR provides a valuable observational constraint on how rotation and convective-core overshoot affect the core evolution of a star.

astro-ph.SR↗

arXiv2018

The White Dwarf Initial-Final Mass Relation for Progenitor Stars From 0.85 to 7.5 M$_\odot$

We present the initial-final mass relation (IFMR) based on the self-consistent analysis of Sirius B and 79 white dwarfs from 13 star clusters. We have also acquired additional signal on eight white dwarfs previously analyzed in the NGC 2099 cluster field, four of which are consistent with membership. These reobserved white dwarfs have masses ranging from 0.72 to 0.97 M$_\odot$, with initial masses from 3.0 to 3.65 M$_\odot$, where the IFMR has an important change in slope that these new data help to observationally confirm. In total, this directly measured IFMR has small scatter ($σ$ = 0.06 M$_\odot$) and spans from progenitors of 0.85 to 7.5 M$_\odot$. Applying two different stellar evolutionary models to infer two different sets of white dwarf progenitor masses shows that when the same model is also used to derive the cluster ages, the resulting IFMR has weak sensitivity to the adopted model at all but the highest initial masses ($>$5.5 M$_\odot$). The non-linearity of the IFMR is also clearly observed with moderate slopes at lower masses (0.08 M$_{\rm final}$/M$_{\rm initial}$) and higher masses (0.11 M$_{\rm final}$/M$_{\rm initial}$) that are broken up by a steep slope (0.19 M$_{\rm final}$/M$_{\rm initial}$) between progenitors from 2.85 to 3.6 M$_\odot$. This IFMR shows total stellar mass loss ranges from 33\% of M$_{\rm initial}$ at 0.83 M$_\odot$ to 83\% of M$_{\rm initial}$ at 7.5 M$_\odot$. Testing this total mass loss for dependence on progenitor metallicity, however, finds no detectable sensitivity across the moderate range of -0.15 $<$ [Fe/H] $<$ +0.15.

astro-ph.SR↗

arXiv2018

Age Determinations of the Hyades, Praesepe, and Pleiades via MESA Models with Rotation

The Hyades, Praesepe, and Pleiades are well studied stellar clusters that anchor important secondary stellar age indicators. Recent studies have shown that main sequence turn off-based ages for these clusters may depend on the degree of rotation in the underlying stellar models. Rotation induces structural instabilities that can enhance the chemical mixing of a star, extending its fuel supply. In addition, rotation introduces a modulation of the star's observed magnitude and color due to the effects of gravity darkening. We aim to investigate the extent to which stellar rotation affects the age determination of star clusters. We utilize the MESA stellar evolution code to create models that cover a range of rotation rates corresponding to $Ω/Ω_c=0.0$ to $0.6$ in $0.1$ dex steps, allowing the assessment of variations in this dimension. The statistical analysis package, MATCH, is employed to derive ages and metallicities by fitting our MESA models to Tycho $B_T$, $V_T$ and 2MASS $J$, $K_s$ color-magnitude diagrams. We find that the derived ages are relatively insensitive to the effects of rotation. For the Hyades, Praesepe, and Pleiades, we derive ages based on synthetic populations that model a distribution of rotation rates or a fixed rate. Across each case, derived ages tend to agree roughly within errors, near $680$, $590$, and $110-160$ Myr for the Hyades, Praesepe, and Pleiades, respectively. These ages are in agreement with Li depletion boundary-based ages and previous analyses that used non-rotating isochrones. Our methods do not provide a strong constraint on the metallicities of these clusters.

astro-ph.SR↗

arXiv2018

Star Cluster Ages in the Gaia Era

We use the framework developed as part of the MESA Isochrones and Stellar Tracks (MIST) project to assess the utility of several types of observables in jointly measuring the age and 1D stellar model parameters in star clusters. We begin with a pedagogical overview summarizing the effects of stellar model parameters, such as the helium abundance, mass-loss efficiency, and the mixing length parameter, on observational diagnostics including the color-magnitude diagram, mass-radius relation, and surface abundances, amongst others. We find that these parameters and the stellar age influence observables in qualitatively distinctive, degeneracy-breaking ways. To assess the current state of affairs, we use the recent Gaia Data Release 2 (DR2) along with data from the literature to investigate three well-studied old open clusters---NGC6819, M67, NGC6791---as case studies. Although there is no obvious tension between the existing observations and the MIST models for NGC6819, there are interesting discrepancies in the cases of M67 and NGC6791. At this time, parallax zero point uncertainties in Gaia DR2 remain one of the limiting factors in the analysis of these clusters. With a combination of exquisite photometry, parallax distances, and cluster memberships from Gaia at the end of its mission, we anticipate precise and accurate ages for these and other star clusters in the Galaxy.

astro-ph.SR↗

arXiv2018

On the Red Giant Branch: Ambiguity in the Surface Boundary Condition Leads to ~100 K Uncertainty in Model Effective Temperatures

The effective temperature (Teff) distribution of stellar evolution models along the red giant branch (RGB) is sensitive to a number of parameters including the overall metallicity, elemental abundance patterns, the efficiency of convection, and the treatment of the surface boundary condition. Recently there has been interest in using observational estimates of the RGB Teff to place constraints on the mixing length parameter, a_MLT, and possible variation with metallicity. Here we use 1D MESA stellar evolution models to explore the sensitivity of the RGB Teff to the treatment of the surface boundary condition. We find that different surface boundary conditions can lead to +/- 100 K metallicity-dependent offsets on the RGB relative to one another in spite of the fact that all models can reproduce the properties of the Sun. Moreover, for a given atmosphere T-tau relation, we find that the RGB Teff is also sensitive to the optical depth at which the surface boundary condition is applied in the stellar model. Nearly all models adopt the photosphere as the location of the surface boundary condition but this choice is somewhat arbitrary. We compare our models to stellar parameters derived from the APOGEE-Kepler sample of first ascent red giants and find that systematic uncertainties in the models due to treatment of the surface boundary condition place a limit of ~100 K below which it is not possible to make firm conclusions regarding the fidelity of the current generation of stellar models.

astro-ph.SR↗

arXiv2017

A precessing Be disk as a possible model for occultation events in GX 304-1

We report on the RXTE detection of a sudden increase in the absorption column density, $N_\mathrm{H}$, during the 2011 May outburst of GX 304-1. The $N_\mathrm{H}$ increased up to ${\sim}16\times 10^{22}$ atoms cm$^{-2}$, which is a factor of 3-4 larger than what is usually measured during the outbursts of GX 304-1 as covered by RXTE. Additionally, an increase in the variability of the hardness ratio as calculated from the energy resolved RXTE-PCA light curves is measured during this time range. We interpret these facts as an occultation event of the neutron star by material in the line of sight. Using a simple 3D model of an inclined and precessing Be disk around the Be type companion, we are able to qualitatively explain the $N_\mathrm{H}$ evolution over time. We are able to constrain the Be-disk density to be on the order of $10^{-11}$ g cm$^{-3}$. Our model strengthens the idea of inclined Be disks as origin of double-peaked outbursts as the derived geometry allows accretion twice per orbit under certain conditions.

astro-ph.HE↗

arXiv2017

The Evolution and Properties of Rotating Massive Star Populations

We investigate the integrated properties of massive (>10 Msun), rotating, single-star stellar populations for a variety of initial rotation rates (v/vcrit=0.0, 0.2, 0.4, 0.5, and 0.6). We couple the new MESA Isochrone and Stellar Tracks (MIST) models to the Flexible Stellar Population Synthesis (FSPS) package, extending the stellar population synthesis models to include the contributions from very massive stars (>100 Msun), which can be significant in the first ~4 Myr after a starburst. These models predict ionizing luminosities that are consistent with recent observations of young nuclear star clusters. We also construct composite stellar populations assuming a distribution of initial rotation rates. Even in low-metallicity environments where rotation has a significant effect on the evolution of massive stars, we find that stellar population models require a significant contribution from fast-rotating (v/vcrit>0.4) stars in order to sustain the production of ionizing photons beyond a few Myr following a starburst. These results have potentially important implications for cosmic reionization by massive stars and the interpretation of nebular emission lines in high-redshift star-forming galaxies.

astro-ph.SR↗