Searcharxiv⌕ Search

arXiv subjects

Kijung Lee

Publications and source records attributed to Kijung Lee.

At least 19 recordsLinked to original sources
arXiv2025

Signal, Noise, and Burnout: A Human-Information Interaction Analysis of Voter Verification in a High-Volatility Environment

The 2024 U.S. Presidential Election unfolded within an information environment of unprecedented volatility, challenging citizens to navigate a torrent of rapidly evolving, often contradictory information while determining what to believe. This study investigates the cognitive mechanisms underlying epistemic self-efficacy - the perceived ability to distinguish accurate news from misinformation - across different information channels during this high-stakes election cycle. Drawing on data from the Pew Research Center's American Trends Panel (Wave 155, September 2024, N = 9,360), we test three hypotheses: (H1) whether reliance on social media predicts lower epistemic self-efficacy compared to mainstream news sources; (H2) whether perceived exposure to inaccurate information mediates this relationship; and (H3) whether information fatigue moderates the cognitive burden of verification across platforms. Contrary to expectations rooted in algorithmic filtering theory, we find no significant differences in reported difficulty determining truth between social media and mainstream news users. Instead, epistemic burden is driven by demographics (age, education) and universal information fatigue, suggesting a "leveling" of the information landscape during periods of extreme volatility. This finding challenges platform-deterministic theories and suggests that interventions to support informed citizenship must address cognitive resilience and attention management rather than platform choice alone.

cs.SI↗
arXiv2025

GranQ: Efficient Channel-wise Quantization via Vectorized Pre-Scaling for Zero-Shot QAT

Zero-shot quantization (ZSQ) enables neural network compression without original training data, making it a promising solution for restricted data access scenarios. To compensate for the lack of data, recent ZSQ methods typically rely on synthetic inputs generated from the full-precision model. However, these synthetic inputs often lead to activation distortion, especially under low-bit settings. To mitigate this, existing methods typically employ per-channel scaling, but they still struggle due to the severe computational overhead during the accumulation process. To overcome this critical bottleneck, we propose GranQ, a novel activation quantization framework that introduces an efficient pre-scaling strategy. Unlike conventional channel-wise methods that repeatedly perform scaling operations during accumulation, GranQ applies scaling factors in a pre-scaling step through fully vectorized computation, eliminating runtime scaling overhead. This design enables GranQ to maintain fine-grained quantization accuracy while significantly reducing computational burden, particularly in low-bit quantization settings. Extensive experiments under quantization-aware training (QAT) settings demonstrate that GranQ consistently outperforms state-of-the-art ZSQ methods across CIFAR and ImageNet. In particular, our method achieves up to 5.45% higher accuracy in the 3-bit setting on CIFAR-100 and even surpasses the full-precision baseline on CIFAR-10.

cs.CV↗
arXiv2025

AnyAnomaly: Zero-Shot Customizable Video Anomaly Detection with LVLM

Video anomaly detection (VAD) is crucial for video analysis and surveillance in computer vision. However, existing VAD models rely on learned normal patterns, which makes them difficult to apply to diverse environments. Consequently, users should retrain models or develop separate AI models for new environments, which requires expertise in machine learning, high-performance hardware, and extensive data collection, limiting the practical usability of VAD. To address these challenges, this study proposes customizable video anomaly detection (C-VAD) technique and the AnyAnomaly model. C-VAD considers user-defined text as an abnormal event and detects frames containing a specified event in a video. We effectively implemented AnyAnomaly using a context-aware visual question answering without fine-tuning the large vision language model. To validate the effectiveness of the proposed model, we constructed C-VAD datasets and demonstrated the superiority of AnyAnomaly. Furthermore, our approach showed competitive results on VAD benchmarks, achieving state-of-the-art performance on UBnormal and UCF-Crime and surpassing other methods in generalization across all datasets. Our code is available online at github.com/SkiddieAhn/Paper-AnyAnomaly.

cs.CV↗
arXiv2024

VideoPatchCore: An Effective Method to Memorize Normality for Video Anomaly Detection

Video anomaly detection (VAD) is a crucial task in video analysis and surveillance within computer vision. Currently, VAD is gaining attention with memory techniques that store the features of normal frames. The stored features are utilized for frame reconstruction, identifying an abnormality when a significant difference exists between the reconstructed and input frames. However, this approach faces several challenges due to the simultaneous optimization required for both the memory and encoder-decoder model. These challenges include increased optimization difficulty, complexity of implementation, and performance variability depending on the memory size. To address these challenges,we propose an effective memory method for VAD, called VideoPatchCore. Inspired by PatchCore, our approach introduces a structure that prioritizes memory optimization and configures three types of memory tailored to the characteristics of video data. This method effectively addresses the limitations of existing memory-based methods, achieving good performance comparable to state-of-the-art methods. Furthermore, our method requires no training and is straightforward to implement, making VAD tasks more accessible. Our code is available online at github.com/SkiddieAhn/Paper-VideoPatchCore.

cs.CV↗
arXiv2023

Examining the impacts of privacy awareness on user's self-disclosure on social media

This research aims to investigate the impact of users' privacy awareness on their self-disclosing behavior. Our primary research question is to investigate how young social media users feel about the benefits and risks of disclosing them-selves on social media and how risk-benefit awareness influences the assess-ment of their self-disclosure. Based on the data we recorded, the factor analysis, and three-way ANOVA, we conclude that users who know more about privacy benefits share more on social media (F= 36.291; df 1; sig < .001) while those who know less about the benefits disclose less on social media. According to the analysis, users who know more about self-disclosure risks share less on so-cial media (F= 7.001; df 1; sig < .001). Users disclose less information on so-cial media platforms based on the different levels of their risk perceptions (df 3, F=.715, sig < 0.5). This indicates that risks on social media platforms vary to some degree. We saw that people's sharing habits based on their levels of risk, benefits, and social media platforms can vary. One thing that remained certain was users' main benefit for engaging and disclosing on social media is their need to stay in touch with friends and their need for community. On the flip side, the main risk was the need not to be impersonated and misunderstood by people. Based on a simple frequency analysis of the open-ended questions we asked In our data collection, the most highlighted words in our responses were "people" and "friends". These were the two main words that stood out in all the data we collected concerning the benefits, risks, and intention to self-disclose.

cs.SI↗
arXiv2023

Why do Tweeters regret sharing? Impacts of Twitter users' perception of sharing risk, perceived problems on Twitter, and the motivation of use on their behavior of regret sharing

This study presents a secondary data analysis of the survey data collected as part of the American Trends Panel series by the Pew Research Center. A logistic regression was performed to ascertain the effects of the perceived risk of sharing, perceived problems on Twitter, and motivation of using Twitter on the likelihood that participants regret sharing on Twitter. The logistic regression model was statistically significant, \c{hi}2(15) = 102.5, p < .001. The model correctly classified 78.5 percent of cases. Whether or not Twitter users regret sharing on Twitter depends on different motivations for using Twitter. We observe that "A way to express my opinion" is statistically significant in the mod-el, indicating that the odds of Twitter users regretting sharing for this motivation is 2.1 times higher than that of entertainment. Perceived risks of potential hostility and visibility were negatively associated with an increased likelihood of regret sharing. In contrast, perceived problems on Twitter concerning misinformation were negatively associated with the likelihood of regret sharing.

cs.SI↗
arXiv2023

Uses and Gratifications of Alternative Social Media: Why do people use Mastodon?

The primary purpose of this investigation is to answer the research questions; 1) What are users' motivations for joining Mastodon?; 2) What are users' gratifications for using Mastodon?; and 3) What are the primary reasons that the users continue to use Mastodon? We analyzed the collected data from the perspective of the Uses and Gratifications Theory. A questionnaire was designed to measure the opinions of Mastodon users from 15 different Mastodon instances. We examined 47 items through exploratory factor analysis using principal components extraction with Varimax with Kaiser Normalization. The results extracted 7 factors of gratification sought (expectation) and 7 factors of gratification obtained. We discovered that the primary reason that the users join and use Mastodon is the ease of controlling and sheltering users' information from data mining. The findings of the gratification sought structure are similar to findings of the gratification obtained structure, and the comparison between the two groups of data suggests that users are satisfied with the ongoing use of Mastodon.

cs.HC↗
arXiv2022

Parabolic Systems with measurable coefficients in weighted Sobolev spaces

In this paper we present a weighted $L_p$-theory of parabolic systems on a half space. The leading coefficients are assumed to be only measurable in $t$ and have small bounded mean oscillations (BMO) with respect to $x$, and the lower order coefficients are allowed to blow up near the boundary.

math.AP↗
arXiv2021

A weighted Sobolev regularity theory of the parabolic equations with measurable coefficients on conic domains in $R^d$

We establish existence, uniqueness, and arbitrary order Sobolev regularity results for the second order parabolic equations with measurable coefficients defined on the conic domains $D$ of the type $$ D(M):=\left\{x\in R^d :\,\frac{x}{|x|}\in M\right\}, \quad \quad M \subset S^{d-1}. $$ We obtain the regularity results by using a system of mixed weights consisting of appropriate powers of the distance to the vertex and of the distance to the boundary. We also provide the sharp ranges of admissible powers of the distance to the vertex and to the boundary.

math.AP↗
arXiv2019

On the regularity of the stochastic heat equation on polygonal domains in $R^2$

We establish existence, uniqueness and higher order weighted $L_p$-Sobolev regularity for the stochastic heat equation with zero Dirichlet boundary condition on angular domains and on polygonal domains in $\mathbb{R}^2$. We use a system of mixed weights consisting of appropriate powers of the distance to the vertexes and of the distance to the boundary to measure the regularity with respect to the space variable. In this way we can capture the influence of both main sources for singularities: the incompatibility between noise and boundary condition on the one hand and the singularities of the boundary on the other hand. The range of admissible powers of the distance to the vertexes is described in terms of the maximal interior angle and is sharp.

math.PR↗
arXiv2016

An $L_p$-estimate for the stochastic heat equation on an angular domain in $\mathbb{R}^2$

We prove a weighted $L_p$-estimate for the stochastic convolution associated to the stochastic heat equation with zero Dirichlet boundary condition on a planar angular domain $\mathcal{D}_{κ_0}\subset\mathbb{R}^2$ with angle $κ_0\in(0,2π)$. Furthermore, we use this estimate to establish existence and uniqueness of a solution to the corresponding equation in suitable weighted $L_p$-Sobolev spaces. In order to capture the singular behaviour of the solution and its derivatives at the vertex, we use powers of the distance to the vertex as weight functions. The admissible range of weight parameters depends explicitly on the angle $κ_0$.

math.PR↗
arXiv2013

On the Lq(Lp)-regularity and Besov smoothness of stochastic parabolic equations on bounded Lipschitz domains

We investigate the regularity of linear stochastic parabolic equations with zero Dirichlet boundary condition on bounded Lipschitz domains $O \subset R^d$ with both theoretical and numerical purpose. We use N.V. Krylov's framework of stochastic parabolic weighted Sobolev spaces $\mathfrak{H}^{\gamma,q}_{p,\theta}(O;T)$. The summability parameters p and q in space and time may differ. Existence and uniqueness of solutions in these spaces is established and the H\"older regularity in time is analysed. Moreover, we prove a general embedding of weighted Lp(O)-Sobolev spaces into the scale of Besov spaces $B^\alpha_{\tau,\tau}(O), 1/\tau=\alpha/d+1/p, \alpha > 0$. This leads to a H\"older-Besov regularity result for the solution process. The regularity in this Besov scale determines the order of convergence that can be achieved by certain nonlinear approximation schemes.

math.PR↗
arXiv2012

A weighted $L_p$-theory for parabolic PDEs with BMO coefficients on $C^1$-domains

In this paper we present a weighted $L_p$-theory of second-order parabolic partial differential equations defined on $C^1$ domains. The leading coefficients are assumed to be measurable in time variable and have VMO (vanishing mean oscillation) or small BMO (bounded mean oscillation) with respect to space variables, and lower order coefficients are allowed to be unbounded and to blow up near the boundary. Our BMO condition is slightly relaxed than the others in the literature.

math.AP↗
arXiv2011

On Initial-Boundary Value Problem of Stochastic Heat Equation in a Lipschitz Cylinder

We consider the initial boundary value problem of non-homogeneous stochastic heat equation. The derivative of the solution with respect to time receives heavy random perturbation. The space boundary is Lipschitz and we impose non-zero cylinder condition. We prove a regularity result after finding suitable spaces for the solution and the pre-assigned datum in the problem. The tools from potential theory, harmonic analysis and probability are used. Some Lemmas are as important as the main Theorem.

math.AP↗
arXiv2011

A $W^n_2$-Theory of Stochastic Parabolic Partial Differential Systems on $C^1$-domains

In this article we present a $W^n_2$-theory of stochastic parabolic partial differential systems. In particular, we focus on non-divergent type. The space domains we consider are $\bR^d$, $\bR^d_+$ and eventually general bounded $C^1$-domains $\mathcal{O}$. By the nature of stochastic parabolic equations we need weighted Sobolev spaces to prove the existence and the uniqueness. In our choice of spaces we allow the derivatives of the solution to blow up near the boundary and moreover the coefficients of the systems are allowed to oscillate to a great extent or blow up near the boundary.

math.PR↗