arXiv subjects

Dae-Won Kim

Publications and source records attributed to Dae-Won Kim.

At least 19 recordsLinked to original sources

arXiv2026

Critical Patch-Aware Sparse Prompting with Decoupled Training for Continual Learning on the Edge

Continual learning (CL) on edge devices requires not only high accuracy but also training-time efficiency to support on-device adaptation under strict memory and computational constraints. While prompt-based continual learning (PCL) is parameter-efficient and achieves competitive accuracy, prior work has focused mainly on accuracy or inference-time performance, often overlooking the memory and computational costs of on-device training. In this paper, we propose CPS-Prompt, a critical patch-aware sparse prompting framework that explicitly targets training-time memory usage and computational cost by integrating critical patch sampling (CPS) for task-aware token reduction and decoupled prompt and classifier training (DPCT) to reduce backpropagation overhead. Experiments on three public benchmarks and real edge hardware show that CPS-Prompt improves peak memory, training time, and energy efficiency by about 1.6x over the balanced CODA-Prompt baseline, while maintaining accuracy within 2% of the state-of-the-art C-Prompt on average and remaining competitive with CODA-Prompt in accuracy. The code is available at https://github.com/laymond1/cps-prompt.

cs.LG↗

arXiv2026

MAND: Modality-Aware Novelty Detection for Open-World Egocentric Activity Recognition

Multimodal egocentric activity recognition integrates visual and inertial cues for robust first-person behavior understanding. However, deploying such systems in open-world environments requires detecting novel activities while continuously learning from non-stationary data streams. Existing methods rely on the main fused logits for novelty scoring, without fully exploiting the complementary evidence available from individual modalities. Because these logits are often dominated by RGB, cues from other modalities, particularly IMU, remain underutilized, and this imbalance worsens as catastrophic forgetting accumulates. To address this, we propose MAND, a modality-aware framework for multimodal egocentric open-world continual learning. At inference, Modality-aware Adaptive Scoring (MoAS) adaptively adjusts modality contributions using sample-wise reliability and refines novelty scoring with deviation and disagreement penalties. During training, Modality-aware Representation Stabilization Training (MoRST) preserves the discriminative capacity of each modality across tasks through modality-specific heads and modality-wise logit distillation. Experiments on a public multimodal egocentric benchmark show that MAND consistently improves novel activity detection and known-class accuracy while substantially reducing FPR95, indicating more reliable open-world recognition. The source code is available at \href{https://github.com/HyeJeongIm/MAND}{github.com/HyeJeongIm/MAND}.

cs.CV↗

arXiv2026

On Evaluation of Unsupervised Feature Selection for Pattern Classification

Unsupervised feature selection aims to identify a compact subset of features that captures the intrinsic structure of data without supervised label. Most existing studies evaluate the performance of methods using the single-label dataset that can be instantiated by selecting a label from multi-label data while maintaining the original features. Because the chosen label can vary arbitrarily depending on the experimental setting, the superiority among compared methods can be changed with regard to which label happens to be selected. Thus, evaluating unsupervised feature selection methods based solely on single-label accuracy is unreasonable for assessing their true discriminative ability. This study revisits this evaluation paradigm by adopting a multi-label classification framework. Experiments on 21 multi-label datasets using several representative methods demonstrate that performance rankings differ markedly from those reported under single-label settings, suggesting the possibility of multi-label evaluation settings for fair and reliable comparison of unsupervised feature selection methods.

cs.LG↗

arXiv2024

PKS 1424-418: A persistent candidate source of the mm$-\gamma$-ray connection?

We present a long-term strong correlation between millimeter (mm) radio and $\gamma$-ray emission in the flat-spectrum radio quasar (FSRQ) PKS 1424-418. The mm$-\gamma$-ray connection in blazars is generally thought to originate from the relativistic jet close to the central engine. We confirm a unique long-lasting mm$-\gamma$-ray correlation of PKS 1424-418 by using detailed correlation analyses and statistical tests, and we find its physical meaning in the source. We employed ~8.5 yr of (sub)mm and $\gamma$-ray light curves observed by ALMA and Fermi-LAT, respectively. From linear and cross-correlation analyses between the light curves, we found a significant, strong mm$-\gamma$-ray correlation over the whole period. We did not find any notable time delay within the uncertainties for the mm$-\gamma$-ray correlation, which means zero lag. The mm wave spectral index values (S$_{\nu}$ $\propto$ $\nu_{\alpha}$) between the band 3 and 7 flux densities indicate a time-variable opacity of the source at (sub)mm wavelengths. Interestingly, the mm wave spectral index becomes temporarily flatter (i.e., $\alpha$ > $-$0.5) when the source flares in the $\gamma$-rays. We relate our results with the jet of PKS 1424-418, and we discuss the origin of the $\gamma$-rays and opacity of the inner (sub)parsec-scale jet regions.

astro-ph.HE↗

arXiv2024

Estimation of Distribution Algorithms with Matrix Transpose in Bayesian Learning

Estimation of distribution algorithms (EDAs) constitute a new branch of evolutionary optimization algorithms, providing effective and efficient optimization performance in a variety of research areas. Recent studies have proposed new EDAs that employ mutation operators in standard EDAs to increase the population diversity. We present a new mutation operator, a matrix transpose, specifically designed for Bayesian structure learning, and we evaluate its performance in Bayesian structure learning. The results indicate that EDAs with transpose mutation give markedly better performance than conventional EDAs.

cs.NE↗

arXiv2024

A new validity measure for fuzzy c-means clustering

A new cluster validity index is proposed for fuzzy clusters obtained from fuzzy c-means algorithm. The proposed validity index exploits inter-cluster proximity between fuzzy clusters. Inter-cluster proximity is used to measure the degree of overlap between clusters. A low proximity value refers to well-partitioned clusters. The best fuzzy c-partition is obtained by minimizing inter-cluster proximity with respect to c. Well-known data sets are tested to show the effectiveness and reliability of the proposed index.

cs.AI↗

arXiv2024

Fuzzy color model and clustering algorithm for color clustering problem

The research interest of this paper is focused on the efficient clustering task for an arbitrary color data. In order to tackle this problem, we have tried to model the inherent uncertainty and vagueness of color data using fuzzy color model. By taking fuzzy approach to color modeling, we could make a soft decision for the vague regions between neighboring colors. The proposed fuzzy color model defined a three dimensional fuzzy color ball and color membership computation method with two inter-color distances. With the fuzzy color model, we developed a new fuzzy clustering algorithm for an efficient partition of color data. Each fuzzy cluster set has a cluster prototype which is represented by fuzzy color centroid.

cs.CV↗

arXiv2024

On selection of centroids of fuzzy clusters for color classification

A novel initialization method in the fuzzy c-means (FCM) algorithm is proposed for the color clustering problem. Given a set of color points, the proposed initialization extracts dominant colors that are the most vivid and distinguishable colors. Color points closest to the dominant colors are selected as initial centroids in the FCM. To obtain the dominant colors and their closest color points, we introduce reference colors and define a fuzzy membership model between a color point and a reference color.

cs.CV↗

arXiv2023

First GMVA observations with the upgraded NOEMA facility: VLBI imaging of BL Lacertae in a flaring state

We analyze a single-epoch Global mm-VLBI Array (GMVA) observation of the blazar BL Lacertae (BL Lac) at 86 GHz from April 2021. The participation of the upgraded, phased Northern Extended Millimetre Array (NOEMA) adds additional sensitivity to the GMVA, which has facilitated the imaging of BL Lac during an unprecedentedly strong $γ$-ray flare. We aim to explore the nature of the inner subparsec jet of BL Lac and the impact of the NOEMA participation in the observation. For the data reduction, we employed two advanced automatic pipelines: rPICARD for the flux density calibration as well as the model-agnostic signal stabilization and GPCAL for the antenna leakage calibration. The conventional hybrid imaging (CLEAN + amplitude and phase self-calibration) was applied to the calibrated visibilities to generate final VLBI images. We performed a ridge-line analysis and Gaussian model-fits on the final jet image to derive the jet parameters. In our data, the presence of NOEMA improves the image sensitivity by a factor of 2.5. The jet shows a clear wiggling structure within 0.4 mas from the core. Our ridge-line analysis suggests the presence of a helical jet structure (i.e., a sinusoidal pattern). Six circular Gaussian components were fitted to the inner jet region. We estimated an apparent brightness temperature of $\sim$3 $\times$ 10$^{12}$ K in the two innermost components. They are likely to be highly boosted by relativistic beaming effect. We find four significant polarized knots in the jet. Interestingly, two of them are located in the core region. Finally, we suggest a number of physical scenarios to interpret our results.

astro-ph.HE↗

arXiv2021

Radio and $γ$-ray activity in the jet of the blazar S5 0716$+$714

We explore the connection between the $γ$-ray and radio emission in the jet of the blazar 0716$+$714 by using 15, 37, and 230 GHz radio and 0.1$-$200 GeV $γ$-ray light curves spanning 10.5 years (2008$-$2019). We find significant positive and negative correlations between radio and $γ$-ray fluxes in different time ranges. The time delays between radio and $γ$-ray emission suggest that the observed $γ$-ray flares originated from multiple regions upstream of the radio core, within a few parsecs from the central engine. Using time-resolved 43 GHz VLBA maps we identified 14 jet components moving downstream along the jet. Their apparent speeds range from 6 to 26 $c$, showing notable variations in their position angles upstream the stationary component ($\sim$0.53 mas from the core). The brightness temperature declines as function of distance from the core according to a power-law which becomes shallower at the location of the stationary component. We also find that the periods at which significant correlations between radio and $γ$-ray emission occur overlap with the times when the jet was oriented to the north. Our results indicate that the passage of a propagating disturbance (or shock) through the radio core and the orientation of the jet might be responsible for the observed correlation between the radio and $γ$-ray variability. We present a scenario that connects the positive correlation and the unusual anti-correlation by combining the production of a flare and a dip at $γ$-rays by a strong moving shock at different distances from the jet apex.

astro-ph.HE↗

arXiv2021

Deep Transfer Learning for Classification of Variable Sources

Ongoing or upcoming surveys such as Gaia, ZTF, or LSST will observe light-curves of billons or more astronomical sources. This presents new challenges for identifying interesting and important types of variability. Collecting a sufficient number of labelled data for training is difficult, however, especially in the early stages of a new survey. Here we develop a single-band light-curve classifier based on deep neural networks, and use transfer learning to address the training data paucity problem by conveying knowledge from one dataset to another. First we train a neural network on 16 variability features extracted from the light-curves of OGLE and EROS-2 variables. We then optimize this model using a small set (e.g. 5%) of periodic variable light-curves from the ASAS dataset in order to transfer knowledge inferred from OGLE/EROS-2 to a new ASAS classifier. With this we achieve good classification results on ASAS, thereby showing that knowledge can be successfully transferred between datasets. We demonstrate similar transfer learning using Hipparcos and ASAS-SN data. We therefore find that it is not necessary to train a neural network from scratch for every new survey, but rather that transfer learning can be used even when only a small set of labelled data is available in the new survey.

astro-ph.IM↗

arXiv2020

Interferometric Monitoring of Gamma-ray Bright AGNs: OJ 287

We present the results of simultaneous multi-frequency imaging observations at 22, 43, 86, and 129\,GHz of OJ\,287. We used the Korean VLBI Network as part of the Interferometric MOnitoring of GAmma-ray Bright active galactic nuclei (iMOGABA). The iMOGABA observations were performed during 31 epochs from 2013 January 16 to 2016 December 28. We also used 15\,GHz OVRO and 225\,GHz SMA flux density data. We analyzed four flux enhancements in the light curves. The estimated time scales of three flux enhancements were similar with time scales of $\sim$50 days at two frequencies. A fourth flux enhancement had a variability timescale approximately twice as long. We found that 225\,GHz enhancements led the 15\,GHz enhancements by a range of 7 to 30 days in the time delay analysis. We found the fractional variability did not change with frequency between 43 and 86\,GHz. We could reliably measure the turnover frequency, $ν_{\rm c}$, of the core of the source in three epochs. This was measured to be in a range from 27 to 50\,GHz and a flux density at the turnover frequency, $S_{\rm m}$, ranging from 3-6\,Jy. The derived SSA magnetic fields, $B_{\rm SSA}$, are in a range from $0.157\pm0.104$ to $0.255\pm0.146$ mG. We estimated the equipartition magnetic field strengths to be in a range from $0.95\pm0.15$ to $1.93\pm0.30$ mG. The equipartition magnetic field strengths are up to a factor of 10 higher than the values of $B_{\rm SSA}$. We conclude that the downstream jet may be more particle energy dominated.

astro-ph.GA↗

arXiv2020

Investigating the connection between gamma-ray activity and relativistic jet in 3C273 during 2015-2019

Due to its powerful radiation over the entire electromagnetic spectrum and its radio jet activity, the blazar 3C273 offers the opportunity to study the physics of $γ$-ray emission from active galactic nuclei. Since a historically strong outburst in 2009, 3C273 showed relatively weak emission in the gamma-ray band over multiple years. However, recent Fermi-Large Area Telescope observations indicate elevated activity during 2015-2019. We aim at constraining the origin of the gamma-ray outbursts towards 3C273 and investigate their connection to the parsec-scale jet. We generate Fermi-LAT gamma-ray light curves with multiple binning intervals and study the spectral properties of the gamma-ray emission. Using a 3-mm ALMA light curve, we study the correlation between radio and gamma-ray emission. Relevant activity in the parsec-scale jet of 3C273 is investigated with 7-mm VLBA observations obtained close in time to notable gamma-ray outbursts. We find two prominent gamma-ray outbursts in 2016 (MJD 57382) and 2017 (MJD 57883) accompanied by mm-wavelength flaring activity. The gamma-ray photon index time series show a weak hump-like feature around the gamma-ray outbursts. The monthly gamma-ray flux-index plot indicates a transition from softer-when-brighter to harder-when-brighter at $1.03\times10^{-7}\rm\,ph\,cm^{-2}\,s^{-1}$. A significant correlation between the gamma-ray and mm-wavelength emission is found, with the radio lagging the gamma-rays by about 105-112 days. The 43-GHz jet images reveal the known stationary features (i.e., the core, S1, and S2) in a region upstream of the jet. We find indication for a propagating disturbance and a polarized knot between the stationary components around the times of both gamma-ray outbursts.

astro-ph.HE↗

arXiv2019

Origin and Evolution of the Multi-band Variability in the Flat Spectrum Radio Source 4C 38.41

The flat spectrum radio quasar 4C 38.41 showed a significant increase of its radio flux density during the period 2012 March - 2015 August which correlates with gamma-ray flaring activity. Multi-frequency simultaneous VLBI observations were conducted as part of the interferometric monitoring of gamma-ray bright active galactic nuclei (iMOGABA) program and supplemented with additional monitoring observations at various bands across the electromagnetic spectrum. The epochs of the maxima for the two largest gamma-ray flares coincide with the ejection of two respective new VLBI components and the evolution of the physical properties seem to be in agreement with the shock-in-jet model. Derived synchrotron self absorption magnetic fields, of the order of 0.1 mG, do not seem to dramatically change during the flares, and are much smaller, by a factor 10,000, than the estimated equipartition magnetic fields, indicating that the source of the flare may be associated with a particle dominated emitting region.

astro-ph.GA↗

arXiv2019

Ejection of Double knots from the radio core of PKS 1510--089 during the strong gamma-ray flares in 2015

PKS 1510--089 is a bright and active $γ$-ray source that showed strong and complex $γ$-ray flares in mid-2015 during which the Major Atmospheric Gamma Imaging Cherenkov telescopes detected variable very high energy (VHE; photon energies $>$100 GeV) emission. We present long-term multi-frequency radio, optical, and $γ$-ray light curves of PKS 1510--089 from 2013 to 2018, and results of an analysis of the jet kinematics and linear polarization using 43 GHz Very Long Baseline Array data observed between late 2015 and mid-2017. We find that a strong radio flare trails the $γ$-ray flares in 2015, showing an optically thick spectrum at the beginning and becoming optically thin over time. Two laterally separated knots of emission are observed to emerge from the radio core nearly simultaneously during the $γ$-ray flares. We detect an edge-brightened linear polarization near the core in the active jet state in 2016, similar to the quiescent jet state in 2008--2013. These observations indicate that the $γ$-ray flares may originate from compression of the knots by a standing shock in the core and the jet might consist of multiple complex layers showing time-dependent behavior, rather than of a simple structure of a fast jet spine and a slow jet sheath.

astro-ph.HE↗

arXiv2019

Core-Jet Blending Effects in Active Galactic Nuclei under the Korean VLBI Network View at 43 GHz

A long standing problem in the study of Active Galactic Nuclei (AGNs) is that the observed VLBI core is in fact a blending of the actual AGN core (classically defined by the $τ=1$ surface) and the upstream regions of the jet or optically thin emitting region flows. This blending may cause some biases towards the observables of the core, such as its flux density, size or brightness temperature, which may lead to misleading interpretation of the derived quantities and physics. We study the effects of such blending under the view of the Korean VLBI Network (KVN) for a sample of AGNs at 43 GHz by comparing their observed properties with observations with the Very Large Baseline Array (VLBA). Our results suggest that the observed core sizes are a factor $\sim11$ larger than these of VLBA, which is similar to the factor expected by considering the different resolutions of the two facilities. We suggest the use of this factor to consider blending effects in KVN measurements. Other parameters, such as flux density or brightness temperature, seem to possess a more complicated dependence.

astro-ph.IM↗

arXiv2018

Exploring the Nature of the 2016 γ-ray Emission in the Blazar 1749+096

Recent Fermi-Large Area Telescope (LAT) light curves indicate an active $γ$-ray state spanning about five months from 2016 June to 2016 October in the BL Lac object 1749+096 (OT 081). During this period, we find two notable $γ$-ray events: an exceptionally strong outburst followed by a significant enhancement (local peak). In this study, we analyze multi-waveband light curves (radio, optical, X-ray, and $γ$-ray) plus very-long baseline interferometry (VLBI) data to investigate the nature of the $γ$-ray events. The $γ$-ray outburst coincides with flux maxima at longer wavelengths. We find a spectral hardening of the $γ$-ray photon index during the $γ$-ray outburst. The photon index shows a transition from a softer-when-brighter to a harder-when-brighter trend at around 1.8 $\times$ $10^{-7}$ ph cm$^{-2}$ s$^{-1}$. We see indication that both the $γ$-ray outburst and the subsequent enhancement precede the propagation of a polarized knot in a region near the VLBI core. The highest polarized intensity, 230\,mJy, and an electric vector position angle rotation, by $\sim$32$^{\circ}$, are detected about 12 days after the $γ$-ray outburst. We conclude that both $γ$-ray events are caused by the propagation of a disturbance in the mm-wave core.

astro-ph.GA↗

arXiv2018

Revealing the Nature of Blazar Radio Cores through Multi-Frequency Polarization Observations with the Korean VLBI Network

We study the linear polarization of the radio cores of eight blazars simultaneously at 22, 43, and 86 GHz with observations obtained by the Korean VLBI Network (KVN) in three epochs between late 2016 and early 2017 in the frame of the Plasma-physics of Active Galactic Nuclei (PAGaN) project. We investigate the Faraday rotation measure (RM) of the cores; the RM is expected to increase with observing frequency if core positions depend on frequency due to synchrotron self-absorption. We find a systematic increase of RMs at higher observing frequencies in our targets. The RM--$ν$ relations follow power-laws with indices distributed around 2, indicating conically expanding outflows serving as Faraday rotating media. Comparing our KVN data with contemporaneous optical polarization data from the Steward Observatory for a few sources, we find indication that the increase of RM with frequency saturates at frequencies of a few hundreds GHz. This suggests that blazar cores are physical structures rather than simple $τ=1$ surfaces. A single region, e.g. a recollimation shock, might dominate the jet emission downstream of the jet launching region. We detect a sign change in the observed RMs of CTA 102 on a time scale of $\approx$1 month, which might be related to new superluminal components emerging from its core undergoing acceleration/deceleration and/or bending. We see indication for quasars having higher core RMs than BL Lac objects, which could be due to denser inflows/outflows in quasars.

astro-ph.HE↗