arXiv subjects

Daun Jeong

Publications and source records attributed to Daun Jeong.

11 recordsLinked to original sources

arXiv2025

S^4M: Boosting Semi-Supervised Instance Segmentation with SAM

Semi-supervised instance segmentation poses challenges due to limited labeled data, causing difficulties in accurately localizing distinct object instances. Current teacher-student frameworks still suffer from performance constraints due to unreliable pseudo-label quality stemming from limited labeled data. While the Segment Anything Model (SAM) offers robust segmentation capabilities at various granularities, directly applying SAM to this task introduces challenges such as class-agnostic predictions and potential over-segmentation. To address these complexities, we carefully integrate SAM into the semi-supervised instance segmentation framework, developing a novel distillation method that effectively captures the precise localization capabilities of SAM without compromising semantic recognition. Furthermore, we incorporate pseudo-label refinement as well as a specialized data augmentation with the refined pseudo-labels, resulting in superior performance. We establish state-of-the-art performance, and provide comprehensive experiments and ablation studies to validate the effectiveness of our proposed approach.

cs.CV↗

arXiv2024

SPACE: SPAtial-aware Consistency rEgularization for anomaly detection in Industrial applications

In this paper, we propose SPACE, a novel anomaly detection methodology that integrates a Feature Encoder (FE) into the structure of the Student-Teacher method. The proposed method has two key elements: Spatial Consistency regularization Loss (SCL) and Feature converter Module (FM). SCL prevents overfitting in student models by avoiding excessive imitation of the teacher model. Simultaneously, it facilitates the expansion of normal data features by steering clear of abnormal areas generated through data augmentation. This dual functionality ensures a robust boundary between normal and abnormal data. The FM prevents the learning of ambiguous information from the FE. This protects the learned features and enables more effective detection of structural and logical anomalies. Through these elements, SPACE is available to minimize the influence of the FE while integrating various data augmentations.In this study, we evaluated the proposed method on the MVTec LOCO, MVTec AD, and VisA datasets. Experimental results, through qualitative evaluation, demonstrate the superiority of detection and efficiency of each module compared to state-of-the-art methods.

cs.CV↗

arXiv2024

Very-Large-Scale GPU-Accelerated Nuclear Gradient of Time-Dependent Density Functional Theory with Tamm-Dancoff Approximation and Range-Separated Hybrid Functionals

Modern graphics processing units (GPUs) provide an unprecedented level of computing power. In this study, we present a high-performance, multi-GPU implementation of the analytical nuclear gradient for Kohn-Sham time-dependent density functional theory (TDDFT), employing the Tamm-Dancoff approximation (TDA) and Gaussian-type atomic orbitals as basis functions. We discuss GPU-efficient algorithms for the derivatives of electron repulsion integrals and exchange-correlation functionals within the range-separated scheme. As an illustrative example, we calculated the TDA-TDDFT gradient of the S1 state of a full-scale green fluorescent protein with explicit water solvent molecules, totaling 4353 atoms, at the wB97X/def2-SVP level of theory. Our algorithm demonstrates favorable parallel efficiencies on a high-speed distributed system equipped with 256 Nvidia A100 GPUs, achieving >70% with up to 64 GPUs and 31% with 256 GPUs, effectively leveraging the capabilities of modern high-performance computing systems.

physics.chem-ph↗

arXiv2024

AI-KD: Adversarial learning and Implicit regularization for self-Knowledge Distillation

We present a novel adversarial penalized self-knowledge distillation method, named adversarial learning and implicit regularization for self-knowledge distillation (AI-KD), which regularizes the training procedure by adversarial learning and implicit distillations. Our model not only distills the deterministic and progressive knowledge which are from the pre-trained and previous epoch predictive probabilities but also transfers the knowledge of the deterministic predictive distributions using adversarial learning. The motivation is that the self-knowledge distillation methods regularize the predictive probabilities with soft targets, but the exact distributions may be hard to predict. Our method deploys a discriminator to distinguish the distributions between the pre-trained and student models while the student model is trained to fool the discriminator in the trained procedure. Thus, the student model not only can learn the pre-trained model's predictive probabilities but also align the distributions between the pre-trained and student models. We demonstrate the effectiveness of the proposed method with network architectures on multiple datasets and show the proposed method achieves better performance than state-of-the-art methods.

cs.CV↗

arXiv2023

Proxy Anchor-based Unsupervised Learning for Continuous Generalized Category Discovery

Recent advances in deep learning have significantly improved the performance of various computer vision applications. However, discovering novel categories in an incremental learning scenario remains a challenging problem due to the lack of prior knowledge about the number and nature of new categories. Existing methods for novel category discovery are limited by their reliance on labeled datasets and prior knowledge about the number of novel categories and the proportion of novel samples in the batch. To address the limitations and more accurately reflect real-world scenarios, in this paper, we propose a novel unsupervised class incremental learning approach for discovering novel categories on unlabeled sets without prior knowledge. The proposed method fine-tunes the feature extractor and proxy anchors on labeled sets, then splits samples into old and novel categories and clusters on the unlabeled dataset. Furthermore, the proxy anchors-based exemplar generates representative category vectors to mitigate catastrophic forgetting. Experimental results demonstrate that our proposed approach outperforms the state-of-the-art methods on fine-grained datasets under real-world scenarios.

cs.CV↗

arXiv2012

Reconstructing Equilibrium Entropy and Enthalpy Profiles from Non-equilibrium Pulling

The Jarzynski identity can be applied to instances when a microscopic system is pulled repeatedly but quickly along some coordinate, allowing the calculation of an equilibrium free energy profile along the pulling coordinate from a set of independent non-equilibrium trajectories. Using the formalism of Wiener stochastic path integrals in which we assign temperature-dependent weights to Langevin trajectories, we derive exact formulae for the temperature derivatives of the free energy profile. This leads naturally to analytical expressions for decomposing a free energy profile into equilibrium entropy and internal energy profiles from non-equilibrium pulling. This decomposition can be done from trajectories evolved at a unique temperature without repeating the measurement as done in finite-difference decompositions. Three distinct analytical expressions for the entropy-energy decomposition are derived: using a time-dependent generalization of the weighted histogram analysis method, a quasi harmonic spring limit, and a Feynman-Kac formula. The three novel formulae of reconstructing the pair of entropy-energy profiles are exemplified by Langevin simulations of a two-dimensional model system prototypical for force-induced biomolecular conformational changes. Connections to single-molecule experimental means to probe the functionals needed in the decomposition are suggested.

cond-mat.stat-mech↗

arXiv2010

Fragility, Stokes-Einstein violation, and correlated local excitations in a coarse-grained model of an ionic liquid

Dynamics of a coarse-grained model for the room-temperature ionic liquid, 1-ethyl-3-methylimidazolium hexafluorophosphate, couched in the united-atom site representation are studied via molecular dynamics simulations. The dynamically heterogeneous behavior of the model resembles that of fragile supercooled liquids. At or close to room temperature, the model ionic liquid exhibits slow dynamics, characterized by nonexponential structural relaxation and subdiffusive behavior. The structural relaxation time, closely related to the viscosity, shows a super-Arrhenius behavior. Local excitations, defined as displacement of an ion exceeding a threshold distance, are found to be mainly responsible for structural relaxation in the alternating structure of cations and anions. As the temperature is lowered, excitations become progressively more correlated. This results in the decoupling of exchange and persistence times, reflecting a violation of the Stokes-Einstein relation.

cond-mat.soft↗

arXiv2008

1/f spectrum and memory function analysis of solvation dynamics in a room-temperature ionic liquid

To understand the non-exponential relaxation associated with solvation dynamics in the ionic liquid 1-ethyl-3-methylimidazolium hexafluorophosphate, we study power spectra of the fluctuating Franck-Condon energy gap of a diatomic probe solute via molecular dynamics simulations. Results show 1/f dependence in a wide frequency range over 2 to 3 decades, indicating distributed relaxation times. We analyze the memory function and solvation time in the framework of the generalized Langevin equation using a simple model description for the power spectrum. It is found that the crossover frequency toward the white noise plateau is directly related to the time scale for the memory function and thus the solvation time. Specifically, the low crossover frequency observed in the ionic liquid leads to a slowly-decaying tail in its memory function and long solvation time. By contrast, acetonitrile characterized by a high crossover frequency and (near) absence of 1/f behavior in its power spectra shows fast relaxation of the memory function and single-exponential decay of solvation dynamics in the long-time regime.

cond-mat.stat-mech↗

arXiv2007

Construction of equilibrium networks with an energy function

We construct equilibrium networks by introducing an energy function depending on the degree of each node as well as the product of neighboring degrees. With this topological energy function, networks constitute a canonical ensemble, which follows the Boltzmann distribution for given temperature. It is observed that the system undergoes a topological phase transition from a random network to a star or a fully-connected network as the temperature is lowered. Both mean-field analysis and numerical simulations reveal strong first-order phase transitions at temperatures which decrease logarithmically with the system size. Quantitative discrepancies of the simulation results from the mean-field prediction are discussed in view of the strong first-order nature.

cond-mat.stat-mech↗

arXiv2005

Slow relaxation in the Ising model on a small-world network with strong long-range interactions

We consider the Ising model on a small-world network, where the long-range interaction strength $J_2$ is in general different from the local interaction strength $J_1$, and examine its relaxation behaviors as well as phase transitions. As $J_2/J_1$ is raised from zero, the critical temperature also increases, manifesting contributions of long-range interactions to ordering. However, it becomes saturated eventually at large values of $J_2/J_1$ and the system is found to display very slow relaxation, revealing that ordering dynamics is inhibited rather than facilitated by strong long-range interactions. To circumvent this problem, we propose a modified updating algorithm in Monte Carlo simulations, assisting the system to reach equilibrium quickly.

cond-mat.stat-mech↗

arXiv2003

Phase transition in the Ising model on a small-world network with distance-dependent interactions

We study the collective behavior of an Ising system on a small-world network with the interaction $J(r) \propto r^{-α}$, where $r$ represents the Euclidean distance between two nodes. In the case of $α= 0$ corresponding to the uniform interaction, the system is known to possess a phase transition of the mean-field nature, while the system with the short-range interaction $(α\to\infty)$ does not exhibit long-range order at any finite temperature. Monte Carlo simulations are performed at various values of $α$, and the critical value $α_c$ beyond which the long-range order does not emerge is estimated to be zero. Thus concluded is the absence of a phase transition in the system with the algebraically decaying interaction $r^{-α}$ for any nonzero positive value of $α$.

cond-mat.stat-mech↗