arXiv subjects
Publications and source records attributed to Minseok Kim.
This paper presents the DaG LLM (David and Goliath Large Language Model), a language model specialized for Korean and fine-tuned through Instruction Tuning across 41 tasks within 13 distinct categories.
Understanding the interplay between charge, nematic, and structural ordering tendencies in cuprate superconductors is critical to unraveling their complex phase diagram. Using pump-probe time-resolved resonant x-ray scattering on the (0 0 1) Bragg peak at the Cu $L_3$ and O $K$ resonances, we investigate non-equilibrium dynamics of $Q_a = Q_b = 0$ nematic order and its association with both charge density wave (CDW) order and lattice dynamics in La$_{1.65}$Eu$_{0.2}$Sr$_{0.15}$CuO$_4$. The orbital selectivity of the resonant x-ray scattering cross-section allows nematicity dynamics associated with the planar O 2$p$ and Cu 3$d$ states to be distinguished from the response of anisotropic lattice distortions. A direct time-domain comparison of CDW translational-symmetry breaking and nematic rotational-symmetry breaking reveals that these broken symmetries remain closely linked in the photoexcited state, consistent with the stability of CDW topological defects in the investigated pump fluence regime.
In this report, we present our award-winning solutions for the Music Demixing Track of Sound Demixing Challenge 2023. First, we propose TFC-TDF-UNet v3, a time-efficient music source separation model that achieves state-of-the-art results on the MUSDB benchmark. We then give full details regarding our solutions for each Leaderboard, including a loss masking approach for noise-robust training. Code for reproducing model training and final submissions is available at github.com/kuielab/sdx23.
Understanding melting in metals is a hot topic of present research. This may be accomplished by pumping the system with infrared (IR) laser radiation, and probing it with hard X-rays produced by an X-ray Free-Electron Laser (XFEL). In this work we studied nanostructured polycrystalline thin films of platinum that were illuminated by IR radiation of increasing fluences. We characterized the structural response as well as the nucleation and propagation of the liquid phase as a function of time delay between the IR pump and X-ray probe. We observed partial melting of the samples for IR fluences higher than 200 mJ$\cdot$cm$^{-2}$. To fit the contribution of the liquid phase to the scattering pattern in platinum we applied a model of liquid metal. The two-temperature model simulations were performed to understand the solid-state fraction of the sample heating process as a function of time delay and fluence.
The phase transition between water and ice is ubiquitous and one of the most important phenomena in nature. Here, we performed time-resolved x-ray scattering experiments capturing the melting and recrystallization dynamics of ice. The ultrafast heating of ice I is induced by an IR laser pulse and probed with an intense x-ray pulse, which provided us with direct structural information on different length scales. From the wide-angle x-ray scattering (WAXS) patterns, the molten fraction, as well as the corresponding temperature at each delay, were determined. The small-angle x-ray scattering (SAXS) patterns, together with the information extracted from the WAXS analysis, provided the time-dependent change of the size and the number of the liquid domains. The results show partial melting (~13 %) and superheating of ice occurring at around 20 ns. After 100 ns, the average size of the liquid domains grows from about 2.5 nm to 4.5 nm by the coalescence of approximately six adjacent domains. Subsequently, we capture the recrystallization of the liquid domains, which occurs on microsecond timescales due to the cooling by heat dissipation and results to a decrease of the average liquid domain size.
Above-bandgap femtosecond optical excitation of a ferroelectric/dielectric BaTiO3/CaTiO3 superlattice leads to structural responses that are a consequence of the screening of the strong electrostatic coupling between the component layers. Time-resolved x-ray free-electron laser diffraction shows that the structural response to optical excitation includes a net lattice expansion of the superlattice consistent with depolarization-field screening driven by the photoexcited charge carriers. The depolarization-field-screening-driven expansion is separate from a photoacoustic pulse launched from the bottom electrode on which the superlattice was epitaxially grown. The distribution of diffracted intensity of superlattice x-ray reflections indicates that the depolarization-field-screening-induced strain includes a photoinduced expansion in the ferroelectric BaTiO3 and a contraction in CaTiO3. The magnitude of expansion in BaTiO3 layers is larger than the contraction in CaTiO3. The difference in the magnitude of depolarization-field-screening-driven strain in the BaTiO3 and CaTiO3 components can arise from the contribution of the oxygen octahedral rotation patterns at the BaTiO3/CaTiO3 interfaces to the polarization of CaTiO3. The depolarization-field-screening-driven polarization reduction in the CaTiO3 layers points to a new direction for the manipulation of polarization in the component layers of a strongly coupled ferroelectric/dielectric superlattice.
We report optical pumping and X-ray absorption spectroscopy experiments at the PAL free electron laser that directly probe the electron dynamics of a graphene monolayer adsorbed on copper in the femtosecond regime. By analyzing the results with ab-initio theory we infer that the excitation of graphene is dominated by indirect excitation from hot electron-hole pairs created in the copper by the optical laser pulse. However, once the excitation is created in graphene, its decay follows a similar path as in many previous studies of graphene adsorbed on semiconductors, i e. rapid excitation of SCOPS (Strongly Coupled Optical Phonons) and eventual thermalization. It is likely that the lifetime of the hot electron-hole pairs in copper governs the lifetime of the electronic excitation of the graphene.
Graph neural networks (GNNs) have achieved remarkable success in recommender systems by representing users and items based on their historical interactions. However, little attention was paid to GNN's vulnerability to exposure bias: users are exposed to a limited number of items so that a system only learns a biased view of user preference to result in suboptimal recommendation quality. Although inverse propensity weighting is known to recognize and alleviate exposure bias, it usually works on the final objective with the model outputs, whereas GNN can also be biased during neighbor aggregation. In this paper, we propose a simple but effective approach, neighbor aggregation via inverse propensity (Navip) for GNNs. Specifically, given a user-item bipartite graph, we first derive propensity score of each user-item interaction in the graph. Then, inverse of the propensity score with Laplacian normalization is applied to debias neighbor aggregation from exposure bias. We validate the effectiveness of our approach through our extensive experiments on two public and Amazon Alexa datasets where the performance enhances up to 14.2%.
Music source separation has been intensively studied in the last decade and tremendous progress with the advent of deep learning could be observed. Evaluation campaigns such as MIREX or SiSEC connected state-of-the-art models and corresponding papers, which can help researchers integrate the best practices into their models. In recent years, the widely used MUSDB18 dataset played an important role in measuring the performance of music source separation. While the dataset made a considerable contribution to the advancement of the field, it is also subject to several biases resulting from a focus on Western pop music and a limited number of mixing engineers being involved. To address these issues, we designed the Music Demixing (MDX) Challenge on a crowd-based machine learning competition platform where the task is to separate stereo songs into four instrument stems (Vocals, Drums, Bass, Other). The main differences compared with the past challenges are 1) the competition is designed to more easily allow machine learning practitioners from other disciplines to participate, 2) evaluation is done on a hidden test set created by music professionals dedicated exclusively to the challenge to assure the transparency of the challenge, i.e., the test set is not accessible from anyone except the challenge organizers, and 3) the dataset provides a wider range of music genres and involved a greater number of mixing engineers. In this paper, we provide the details of the datasets, baselines, evaluation metrics, evaluation results, and technical challenges for future competitions.
Online recommender systems should be always aligned with users' current interest to accurately suggest items that each user would like. Since user interest usually evolves over time, the update strategy should be flexible to quickly catch users' current interest from continuously generated new user-item interactions. Existing update strategies focus either on the importance of each user-item interaction or the learning rate for each recommender parameter, but such one-directional flexibility is insufficient to adapt to varying relationships between interactions and parameters. In this paper, we propose MeLON, a meta-learning based novel online recommender update strategy that supports two-directional flexibility. It is featured with an adaptive learning rate for each parameter-interaction pair for inducing a recommender to quickly learn users' up-to-date interest. The procedure of MeLON is optimized following a meta-learning approach: it learns how a recommender learns to generate the optimal learning rates for future updates. Specifically, MeLON first enriches the meaning of each interaction based on previous interactions and identifies the role of each parameter for the interaction; and then combines these two pieces of information to generate an adaptive learning rate. Theoretical analysis and extensive evaluation on three real-world online recommender datasets validate the effectiveness of MeLON.
Deep learning has achieved remarkable success in numerous domains with help from large amounts of big data. However, the quality of data labels is a concern because of the lack of high-quality labels in many real-world scenarios. As noisy labels severely degrade the generalization performance of deep neural networks, learning from noisy labels (robust training) is becoming an important task in modern deep learning applications. In this survey, we first describe the problem of learning with label noise from a supervised learning perspective. Next, we provide a comprehensive review of 62 state-of-the-art robust training methods, all of which are categorized into five groups according to their methodological difference, followed by a systematic comparison of six properties used to evaluate their superiority. Subsequently, we perform an in-depth analysis of noise rate estimation and summarize the typically used evaluation methodology, including public noisy datasets and evaluation metrics. Finally, we present several promising research directions that can serve as a guideline for future studies. All the contents will be available at https://github.com/songhwanjun/Awesome-Noisy-Labels.
Tin monosulfide (SnS) usually exhibits p-type conduction due to the low formation enthalpy of acceptor-type defects, and as a result n-type SnS thin films have never been obtained. This study realizes n-type conduction in SnS thin films for the first time by using RF-magnetron sputtering with Cl doping and sulfur plasma source during deposition. N-type SnS thin films are obtained at all the substrate temperatures employed in this study (221-341 C), exhibiting carrier concentrations and Hall mobilities of ~2 x 10 18 cm-3 and 0.1-1 cm V-1s-1, respectively. The films prepared without sulfur plasma source, on the other hand, exhibit p-type conduction despite containing a comparable amount of Cl donors. This is likely due to a significant amount of acceptor-type defects originating from sulfur deficiency in p-type films, which appears as a broad optical absorption within the band gap. The demonstration of n-type SnS thin films in this study is a breakthrough for the realization of SnS homojunction solar cells, which are expected to have a higher conversion efficiency than the conventional heterojunction SnS solar cells.
Understanding ultrafast magnetization dynamics on the microscopic level is of strong current interest due to the potential for applications in information storage. In recent years, the spin-lattice coupling has been recognized to be essential for ultrafast magnetization dynamics. Magnetoelectric multiferroics of type II possess intrinsic correlations among magnetic sublattices and electric polarization (P) through spin-lattice coupling, enabling fundamentally coupled dynamics between spins and lattice. Here we report on ultrafast magnetization dynamics in a room-temperature multiferroic hexaferrite possessing ferrimagnetic and antiferromagnetic sublattices, revealed by time-resolved resonant x-ray diffraction. A femtosecond above-bandgap excitation triggers a coherent magnon in which the two magnetic sublattices entangle and give rise to a transient modulation of P. A novel microscopic mechanism for triggering the coherent magnon in this ferrimagnetic insulator based on the spin-lattice coupling is proposed. Our finding opens up a novel but general pathway for ultrafast control of magnetism.
Recently, many methods based on deep learning have been proposed for music source separation. Some state-of-the-art methods have shown that stacking many layers with many skip connections improve the SDR performance. Although such a deep and complex architecture shows outstanding performance, it usually requires numerous computing resources and time for training and evaluation. This paper proposes a two-stream neural network for music demixing, called KUIELab-MDX-Net, which shows a good balance of performance and required resources. The proposed model has a time-frequency branch and a time-domain branch, where each branch separates stems, respectively. It blends results from two streams to generate the final estimation. KUIELab-MDX-Net took second place on leaderboard A and third place on leaderboard B in the Music Demixing Challenge at ISMIR 2021. This paper also summarizes experimental results on another benchmark, MUSDB18. Our source code is available online.
X-ray free electron lasers (XFEL) create femtosecond X-ray pulses with high brightness and high longitudinal coherence allowing to extend X-ray spectroscopy and scattering techniques into the ultrafast time-domain. These X-rays are a powerful probe for studying coherent quasiparticle excitations in condensed matter triggered by an impulsive optical laser pump. However, unlike coherent phonons, other quasiparticles have been rarely observed due to small signal changes and lack of standards for the identification. Here, we exploit resonant magnetic X-ray diffraction using an XFEL to visualize a photoexcited coherent magnon in space and time. Large intensity oscillations in antiferromagnetic and ferromagnetic Bragg reflections from precessing moment are observed in a multiferroic Y-type hexaferrite. The precession trajectory reveals that a large, long-lived, photoinduced magnetic-field changes the net magnetization substantially through the large-amplitude of the magnon. This work demonstrates an efficient XFEL probe for the coherent magnon in the spotlight for opto-spintronics application.
Ultrafast lasers are an increasingly important tool to control and stabilize emergent phases in quantum materials. Among a variety of possible excitation protocols, a particularly intriguing route is the direct light-engineering of microscopic electronic parameters, such as the electron hopping and the local Coulomb repulsion (Hubbard $U$). In this work, we use time-resolved x-ray absorption spectroscopy to demonstrate the light-induced renormalization of the Hubbard $U$ in a cuprate superconductor, La$_{1.905}$Ba$_{0.095}$CuO$_4$. We show that intense femtosecond laser pulses induce a substantial redshift of the upper Hubbard band, while leaving the Zhang-Rice singlet energy unaffected. By comparing the experimental data to time-dependent spectra of single- and three-band Hubbard models, we assign this effect to a $\sim140$ meV reduction of the onsite Coulomb repulsion on the copper sites. Our demonstration of a dynamical Hubbard $U$ renormalization in a copper oxide paves the way to a novel strategy for the manipulation of superconductivity, magnetism, as well as to the realization of other long-range-ordered phases in light-driven quantum materials.
This paper studies a deep learning approach for binary assignment problems in wireless networks, which identifies binary variables for permutation matrices. This poses challenges in designing a structure of a neural network and its training strategies for generating feasible assignment solutions. To this end, this paper develop a new Sinkhorn neural network which learns a non-convex projection task onto a set of permutation matrices. An unsupervised training algorithm is proposed where the Sinkhorn neural network can be applied to network assignment problems. Numerical results demonstrate the effectiveness of the proposed method in various network scenarios.
Real-world data inevitably contains noisy labels, which induce the poor generalization of deep neural networks. It is known that the network typically begins to rapidly memorize false-labeled samples after a certain point of training. Thus, to counter the label noise challenge, we propose a novel self-transitional learning method called MORPH, which automatically switches its learning phase at the transition point from seeding to evolution. In the seeding phase, the network is updated using all the samples to collect a seed of clean samples. Then, in the evolution phase, the network is updated using only the set of arguably clean samples, which precisely keeps expanding by the updated network. Thus, MORPH effectively avoids the overfitting to false-labeled samples throughout the entire training period. Extensive experiments using five real-world or synthetic benchmark datasets demonstrate substantial improvements over state-of-the-art methods in terms of robustness and efficiency.