arXiv subjects

Minjung Kim

Publications and source records attributed to Minjung Kim.

35 records · Page 2Linked to original sources

arXiv2023

Modeling Backward-Angle ($u$-channel) Virtual Compton Scattering at an Electron-Ion Collider

High-energy backward ($u$-channel) reactions can involve very large momentum transfers to the target baryons, shifting them by many units of rapidity. These reactions are difficult to understand in conventional models in which baryon number is carried by the valence quarks. Backward Compton scattering is an especially attractive experimental target, because of its simple final state. There is currently limited data on this process, and that data is at low center-of-mass energies. In this paper, we examine the prospects for studying backward Compton scattering at the future Electron-Ion Collider (EIC). We model the cross-section and kinematics using the limited data on backward Compton scattering and backward meson production, and then simulate Compton scattering at EIC energies, in a simple model of the ePIC detector. Generally, the proton is scattered toward mid-rapidity, while the produced photon is in the far-forward region, visible in a Zero Degree Calorimeter (ZDC). We show that the background from backward $π^0$ production can be rejected using a high-resolution, well-segmented ZDC.

hep-ph↗

arXiv2023

Regular Time-series Generation using SGM

Score-based generative models (SGMs) are generative models that are in the spotlight these days. Time-series frequently occurs in our daily life, e.g., stock data, climate data, and so on. Especially, time-series forecasting and classification are popular research topics in the field of machine learning. SGMs are also known for outperforming other generative models. As a result, we apply SGMs to synthesize time-series data by learning conditional score functions. We propose a conditional score network for the time-series generation domain. Furthermore, we also derive the loss function between the score matching and the denoising score matching in the time-series generation domain. Finally, we achieve state-of-the-art results on real-world datasets in terms of sampling diversity and quality.

cs.LG↗

arXiv2022

Feature Disentanglement Learning with Switching and Aggregation for Video-based Person Re-Identification

In video person re-identification (Re-ID), the network must consistently extract features of the target person from successive frames. Existing methods tend to focus only on how to use temporal information, which often leads to networks being fooled by similar appearances and same backgrounds. In this paper, we propose a Disentanglement and Switching and Aggregation Network (DSANet), which segregates the features representing identity and features based on camera characteristics, and pays more attention to ID information. We also introduce an auxiliary task that utilizes a new pair of features created through switching and aggregation to increase the network's capability for various camera scenarios. Furthermore, we devise a Target Localization Module (TLM) that extracts robust features against a change in the position of the target according to the frame flow and a Frame Weight Generation (FWG) that reflects temporal information in the final representation. Various loss functions for disentanglement learning are designed so that each component of the network can cooperate while satisfactorily performing its own role. Quantitative and qualitative results from extensive experiments demonstrate the superiority of DSANet over state-of-the-art methods on three benchmark datasets.

cs.CV↗

arXiv2022

Occluded Person Re-Identification via Relational Adaptive Feature Correction Learning

Occluded person re-identification (Re-ID) in images captured by multiple cameras is challenging because the target person is occluded by pedestrians or objects, especially in crowded scenes. In addition to the processes performed during holistic person Re-ID, occluded person Re-ID involves the removal of obstacles and the detection of partially visible body parts. Most existing methods utilize the off-the-shelf pose or parsing networks as pseudo labels, which are prone to error. To address these issues, we propose a novel Occlusion Correction Network (OCNet) that corrects features through relational-weight learning and obtains diverse and representative features without using external networks. In addition, we present a simple concept of a center feature in order to provide an intuitive solution to pedestrian occlusion scenarios. Furthermore, we suggest the idea of Separation Loss (SL) for focusing on different parts between global features and part features. We conduct extensive experiments on five challenging benchmark datasets for occluded and holistic Re-ID tasks to demonstrate that our method achieves superior performance to state-of-the-art methods especially on occluded scene.

cs.CV↗

arXiv2022

Tackling Background Distraction in Video Object Segmentation

Semi-supervised video object segmentation (VOS) aims to densely track certain designated objects in videos. One of the main challenges in this task is the existence of background distractors that appear similar to the target objects. We propose three novel strategies to suppress such distractors: 1) a spatio-temporally diversified template construction scheme to obtain generalized properties of the target objects; 2) a learnable distance-scoring function to exclude spatially-distant distractors by exploiting the temporal consistency between two consecutive frames; 3) swap-and-attach augmentation to force each object to have unique features by providing training samples containing entangled objects. On all public benchmark datasets, our model achieves a comparable performance to contemporary state-of-the-art approaches, even with real-time performance. Qualitative results also demonstrate the superiority of our approach over existing methods. We believe our approach will be widely used for future VOS research.

cs.CV↗

arXiv2022

SOS: Score-based Oversampling for Tabular Data

Score-based generative models (SGMs) are a recent breakthrough in generating fake images. SGMs are known to surpass other generative models, e.g., generative adversarial networks (GANs) and variational autoencoders (VAEs). Being inspired by their big success, in this work, we fully customize them for generating fake tabular data. In particular, we are interested in oversampling minor classes since imbalanced classes frequently lead to sub-optimal training outcomes. To our knowledge, we are the first presenting a score-based tabular data oversampling method. Firstly, we re-design our own score network since we have to process tabular data. Secondly, we propose two options for our generation method: the former is equivalent to a style transfer for tabular data and the latter uses the standard generative policy of SGMs. Lastly, we define a fine-tuning method, which further enhances the oversampling quality. In our experiments with 6 datasets and 10 baselines, our method outperforms other oversampling methods in all cases.

cs.LG↗

arXiv2022

3D-GIF: 3D-Controllable Object Generation via Implicit Factorized Representations

While NeRF-based 3D-aware image generation methods enable viewpoint control, limitations still remain to be adopted to various 3D applications. Due to their view-dependent and light-entangled volume representation, the 3D geometry presents unrealistic quality and the color should be re-rendered for every desired viewpoint. To broaden the 3D applicability from 3D-aware image generation to 3D-controllable object generation, we propose the factorized representations which are view-independent and light-disentangled, and training schemes with randomly sampled light conditions. We demonstrate the superiority of our method by visualizing factorized representations, re-lighted images, and albedo-textured meshes. In addition, we show that our approach improves the quality of the generated geometry via visualization and quantitative comparison. To the best of our knowledge, this is the first work that extracts albedo-textured meshes with unposed 2D images without any additional labels or assumptions.

cs.CV↗

arXiv2021

Pixel-Level Bijective Matching for Video Object Segmentation

Semi-supervised video object segmentation (VOS) aims to track the designated objects present in the initial frame of a video at the pixel level. To fully exploit the appearance information of an object, pixel-level feature matching is widely used in VOS. Conventional feature matching runs in a surjective manner, i.e., only the best matches from the query frame to the reference frame are considered. Each location in the query frame refers to the optimal location in the reference frame regardless of how often each reference frame location is referenced. This works well in most cases and is robust against rapid appearance variations, but may cause critical errors when the query frame contains background distractors that look similar to the target object. To mitigate this concern, we introduce a bijective matching mechanism to find the best matches from the query frame to the reference frame and vice versa. Before finding the best matches for the query frame pixels, the optimal matches for the reference frame pixels are first considered to prevent each reference frame pixel from being overly referenced. As this mechanism operates in a strict manner, i.e., pixels are connected if and only if they are the sure matches for each other, it can effectively eliminate background distractors. In addition, we propose a mask embedding module to improve the existing mask propagation method. By embedding multiple historic masks with coordinate information, it can effectively capture the position information of a target object.

cs.CV↗

arXiv2021

Drop-Bottleneck: Learning Discrete Compressed Representation for Noise-Robust Exploration

We propose a novel information bottleneck (IB) method named Drop-Bottleneck, which discretely drops features that are irrelevant to the target variable. Drop-Bottleneck not only enjoys a simple and tractable compression objective but also additionally provides a deterministic compressed representation of the input variable, which is useful for inference tasks that require consistent representation. Moreover, it can jointly learn a feature extractor and select features considering each feature dimension's relevance to the target task, which is unattainable by most neural network-based IB methods. We propose an exploration method based on Drop-Bottleneck for reinforcement learning tasks. In a multitude of noisy and reward sparse maze navigation tasks in VizDoom (Kempka et al., 2016) and DMLab (Beattie et al., 2016), our exploration method achieves state-of-the-art performance. As a new IB framework, we demonstrate that Drop-Bottleneck outperforms Variational Information Bottleneck (VIB) (Alemi et al., 2017) in multiple aspects including adversarial robustness and dimensionality reduction.

cs.LG↗

arXiv2020

Timing Detectors with SiPM read-out for the MUSE Experiment at PSI

The Muon Scattering Experiment at the Paul Scherrer Institut uses a mixed beam of electrons, muons, and pions, necessitating precise timing to identify the beam particles and reactions they cause. We describe the design and performance of three timing detectors using plastic scintillator read out with silicon photomultipliers that have been built for the experiment. The Beam Hodoscope, upstream of the scattering target, counts the beam flux and precisely times beam particles both to identify species and provide a starting time for time-of-flight measurements. The Beam Monitor, downstream of the scattering target, counts the unscattered beam flux, helps identify background in scattering events, and precisely times beam particles for time-of-flight measurements. The Beam Focus Monitor, mounted on the target ladder under the liquid hydrogen target inside the target vacuum chamber, is used in dedicated runs to sample the beam spot at three points near the target center, where the beam should be focused.

physics.ins-det↗

arXiv2020

Boson Slave Solver (BoSS) v1.1

Accurate and computationally efficient modeling of systems of interacting electrons is an outstanding problem in theoretical and computational materials science. For materials where strong electronic interactions are primarily of a localized character and act within a subspace of localized quantum states on separate atomic sites (e.g., in transition metal and rare-earth compounds), their electronic behaviors are typically described by the Hubbard model and its extensions. In this work, we describe BoSS (Boson Slave Solver), a software implementation of the slave-boson method appropriate for describing a variety of extended Hubbard models, namely $p-d$ models that include both the interacting atomic sites ("$d$" states) and non-interacting or ligand sites ("$p$" states). We provide a theoretical background, a description of the equations solved by BoSS, an overview of the algorithms used, the key input/output and control variables of the software program, and tutorial examples of its use featuring band renormalization in SrVO$_3$, Ni $3d$ multiplet structure in LaNiO$_3$, and the relation between the formation of magnetic moments and insulating behavior in SmNiO$_3$. BoSS interfaces directly with popular electronic structure codes: it can read the output of the Wannier90 software package which postprocesses results from workhorse electronic structure software such as Quantum Espresso or VASP.

cond-mat.str-el↗

arXiv2018

Scalable GW software for quasiparticle properties using OpenAtom

The GW method, which can describe accurately electronic excitations, is one of the most widely used ab initio electronic structure technique and allows the physics of both molecular and condensed phase materials to be studied. However, the applications of the GW method to large systems require supercomputers and highly parallelized software to overcome the high computational complexity of the method scaling as $O(N^4)$. Here, we develop efficient massively-parallel GW software for the plane-wave basis set by revisiting the standard GW formulae in order to discern the optimal approaches for each phase of the GW calculation for massively parallel computation. These best numerical practices are implemented into the OpenAtom software which is written on top of charm++ parallel framework. We then evaluate the performance of our new software using range of system sizes. Our GW software shows significantly improved parallel scaling compared to publically available GW software on the Mira and Blue Waters supercomputers, two of largest most powerful platforms in the world.

physics.comp-ph↗

arXiv2018

Memorization Precedes Generation: Learning Unsupervised GANs with Memory Networks

We propose an approach to address two issues that commonly occur during training of unsupervised GANs. First, since GANs use only a continuous latent distribution to embed multiple classes or clusters of data, they often do not correctly handle the structural discontinuity between disparate classes in a latent space. Second, discriminators of GANs easily forget about past generated samples by generators, incurring instability during adversarial training. We argue that these two infamous problems of unsupervised GAN training can be largely alleviated by a learnable memory network to which both generators and discriminators can access. Generators can effectively learn representation of training samples to understand underlying cluster distributions of data, which ease the structure discontinuity problem. At the same time, discriminators can better memorize clusters of previously generated samples, which mitigate the forgetting problem. We propose a novel end-to-end GAN model named memoryGAN, which involves a memory network that is unsupervisedly trainable and integrable to many existing GAN models. With evaluations on multiple datasets such as Fashion-MNIST, CelebA, CIFAR10, and Chairs, we show that our model is probabilistically interpretable, and generates realistic image samples of high visual fidelity. The memoryGAN also achieves the state-of-the-art inception scores over unsupervised GAN models on the CIFAR10 dataset, without any optimization tricks and weaker divergences.

cs.LG↗

arXiv2018

Imaginary time, shredded propagator method for large-scale GW calculations

The GW method is a many-body approach capable of providing quasiparticle bands for realistic systems spanning physics, chemistry, and materials science. Despite its power, GW is not routinely applied to large complex materials due to its computational expense. We perform an exact recasting of the GW polarizability and the self-energy as Laplace integrals over imaginary time propagators. We then "shred" the propagators (via energy windowing) and approximate them in a controlled manner by using Gauss-Laguerre quadrature and discrete variable methods to treat the imaginary time propagators in real space. The resulting cubic scaling GW method has a sufficiently small prefactor to outperform standard quartic scaling methods on small systems (>=10 atoms) and also represents a substantial improvement over several other cubic methods tested. This approach is useful for evaluating quantum mechanical response function involving large sums containing energy (difference) denominators.

cond-mat.mtrl-sci↗

arXiv2016

Determination of the thickness and orientation of few-layer tungsten ditelluride using polarized Raman spectroscopy

Orthorhombic tungsten ditelluride (WTe2), with a distorted 1T structure, exhibits a large magnetoresistance that depends on the orientation, and its electrical characteristics changes rom semimetallic to insulating as the thickness decreases. Through polarized Raman spectroscopy in combination with transmission electron diffraction, we establish a reliable method to determine the thickness and crystallographic orientation of few-layer WTe2. The Raman spectrum shows a pronounced dependence on the polarization of the excitation laser. We found that the separation between two Raman peaks at ~90 cm-1 and at 80-86 cm-1, depending on thickness, is a reliable fingerprint for determination of the thickness. For determination of the crystallographic orientation, the polarization dependence of the A1 modes, measured with the 632.8-nm excitation, turns out to be the most reliable. We also discovered that the polarization behaviors of some of the Raman peaks depend on the excitation wavelength as well as thickness, indicating a close interplay between the band structure and anisotropic Raman scattering cross section.

cond-mat.mes-hall↗

arXiv2015

Photocurrent generation at ABA/ABC lateral junction in tri-layer graphene photodetector

Metal-graphene-metal photodetectors utilize photocurrent generated near the graphene/metal junctions and have many advantages including high speed and broad-band operation. Here, we report on photocurrent generation at ABA/ABC stacking domain junctions in tri-layer graphene with a responsivity of 0.18 A/W. Unlike usual metal-graphene-metal devices, the photocurrent is generated in the middle of the graphene channel, not confined to the vicinity of the metal electrodes. The magnitude and the direction of the photocurrent depend on the back-gate bias. Theoretical calculations show that there is a built-in band offset between the two stacking domains, and the dominant mechanism of the photocurrent is the photo-thermoelectric effect due to the Seebeck coefficient difference.

cond-mat.mes-hall↗

arXiv2012

Polarization dependence of photocurrent in a metal-graphene-metal device

The dependence of the photocurrent generated in a Pd/graphene/Ti junction device on the incident photon polarization is studied. Spatially resolved photocurrent images were obtained as the incident photon polarization is varied. The photocurrent is maximum when the polarization direction is perpendicular to the graphene channel direction and minimum when the two directions are parallel. This polarization dependence can be explained as being due to the anisotropic electron-photon interaction of Dirac electrons in graphene.

cond-mat.mes-hall↗