Sungwon Han

Sungwon Han

Auto-generated profile, based on publicly available information

Sungwon Han

Korea Advanced Institute of Science and Technology, Daejeon, Republic of Korea

Research Focus & Metrics

Academic Career Length

17Years (2010-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
3 November 2022
Verified
Book Chapter
Closed access
PDF file
SHSungwon Han
Sungwon Han
SPSungwon Park
Sungwon Park
FWFangzhao Wu
Fangzhao Wu
SKSundong Kim
Sundong Kim
CWChuhan Wu
Chuhan Wu
MCMeeyoung Cha
Meeyoung Cha

16 March 2021 in Journal of Medical Internet Research
Verified
Journal Article
Open access
SPSungkyu Shaun Park
Sungkyu Shaun Park
SHSungwon Han
Sungwon Han
JKJeongwook Kim
Jeongwook Kim
MMMir Majid Molaie
Mir Majid Molaie
HVHoang Dieu Vu
Hoang Dieu Vu
MCMeeyoung Cha
Meeyoung Cha
Background: COVID-19, caused by SARS-CoV-2, has led to a global pandemic. The World Health Organization has also declared an infodemic (ie, a plethora of information regarding COVID-19 containing both false and accurate information circulated on the internet). Hence, it has become critical to test the veracity of information shared online and analyze the evolution of discussed topics among citizens related to the pandemic. Objective: This research analyzes the public discourse on COVID-19. It characterizes risk communication patterns in four Asian countries with outbreaks at varying degrees of severity: South Korea, Iran, Vietnam, and India. Methods: We collected tweets on COVID-19 from four Asian countries in the early phase of the disease outbreak from January to March 2020. The data set was collected by relevant keywords in each language, as suggested by locals. We present a method to automatically extract a time-topic cohesive relationship in an unsupervised fashion based on natural language processing. The extracted topics were evaluated qualitatively based on their semantic meanings. Results: This research found that each government's official phases of the epidemic were not well aligned with the degree of public attention represented by the daily tweet counts. Inspired by the issue-attention cycle theory, the presented natural language processing model can identify meaningful transition phases in the discussed topics among citizens. The analysis revealed an inverse relationship between the tweet count and topic diversity. Conclusions: This paper compares similarities and differences of pandemic-related social media discourse in Asian countries. We observed multiple prominent peaks in the daily tweet counts across all countries, indicating multiple issue-attention cycles. Our analysis identified which topics the public concentrated on; some of these topics were related to misinformation and hate speech. These findings and the ability to quickly identify key topics can empower global efforts to fight against an infodemic during a pandemic.

5 December 2019 in JMIR mHealth and uHealth
Verified
Journal Article
Open access
SPSungkyu Shaun Park
Sungkyu Shaun Park
SLSang Won Lee
Sang Won Lee
SHSungwon Han
Sungwon Han
MCMeeyoung Cha
Meeyoung Cha
Background: As societies become more complex, larger populations suffer from insomnia. In 2014, the US Centers for Disease Control and Prevention declared that sleep disorders should be dealt with as a public health epidemic. However, it is hard to provide adequate treatment for each insomnia sufferer, since various behavioral characteristics influence symptoms of insomnia collectively. Objective: We aim to develop a neural-net based unsupervised user clustering method towards insomnia sufferers in order to clarify the unique traits for each derived groups. Unlike the current diagnosis of insomnia that requires qualitative analysis from interview results, the classification of individuals with insomnia by using various information modalities from smart bands and neural-nets can provide better insight into insomnia treatments. Methods: This study, as part of the precision psychiatry initiative, is based on a smart band experiment conducted over 6 weeks on individuals with insomnia. During the experiment period, a total of 42 participants (19 male; average age 22.00 [SD 2.79]) from a large university wore smart bands 24/7, and 3 modalities were collected and examined: sleep patterns, daily activities, and personal demographics. We considered the consecutive daily information as a form of images, learned the latent variables of the images via a convolutional autoencoder (CAE), and clustered and labeled the input images based on the derived features. We then converted consecutive daily information into a sequence of the labels for each subject and finally clustered the people with insomnia based on their predominant labels. Results: Our method identified 5 new insomnia-activity clusters of participants that conventional methods have not recognized, and significant differences in sleep and behavioral characteristics were shown among groups (analysis of variance on rank: F4,37=2.36, P=.07 for the sleep_min feature; F4,37=9.05, P<.001 for sleep_efficiency; F4,37=8.16, P<.001 for active_calorie; F4,37=6.53, P<.001 for walks; and F4,37=3.51, P=.02 for stairs). Analyzing the consecutive data through a CAE and clustering could reveal intricate connections between insomnia and various everyday activity markers. Conclusions: Our research suggests that unsupervised learning allows health practitioners to devise precise and tailored interventions at the level of data-guided user clusters (ie, precision psychiatry), which could be a novel solution to treating insomnia and other mental disorders.

26 October 2023 in Nature Communications
Verified
Journal Article
Open access
PDF file
DADonghyun Ahn
Donghyun Ahn
JYJeasurk Yang
Jeasurk Yang
MCMeeyoung Cha
Meeyoung Cha
HYHyunjoo Yang
Hyunjoo Yang
JKJihee Kim
Jihee Kim
SPSungwon Park
Sungwon Park
Machine learning approaches using satellite imagery are providing accessible ways to infer socioeconomic measures without visiting a region. However, many algorithms require integration of ground-truth data, while regional data are scarce or even absent in many countries. Here we present our human-machine collaborative model which predicts grid-level economic development using publicly available satellite imagery and lightweight subjective ranking annotation without any ground data. We applied the model to North Korea and produced fine-grained predictions of economic development for the nation where data is not readily available. Our model suggests substantial development in the country’s capital and areas with state-led development projects in recent years. We showed the broad applicability of our model by examining five of the least developed countries in Asia, covering 400,000 grids. Our method can both yield highly granular economic information on hard-to-visit and low-resource regions and can potentially guide sustainable development programs.

4 August 2023
Verified
Conference Paper
Open access
SPSungwon Park
Sungwon Park
SHSungwon Han
Sungwon Han
FWFangzhao Wu
Fangzhao Wu
SKSundong Kim
Sundong Kim
BZBin Zhu
Bin Zhu
MCMeeyoung Cha
Meeyoung Cha
Federated learning enables learning from decentralized data sources without compromising privacy, which makes it a crucial technique. However, it is vulnerable to model poisoning attacks, where malicious clients interfere with the training process. Previous defense mechanisms have focused on the server-side by using careful model aggregation, but this may not be effective when the data is not identically distributed or when attackers can access the information of benign clients. In this paper, we propose a new defense mechanism that focuses on the client-side, called FedDefender, to help benign clients train robust local models and avoid the adverse impact of malicious model updates from attackers, even when a server-side defense cannot identify or remove adversaries. Our method consists of two main components: (1) attack-tolerant local meta update and (2) attack-tolerant global knowledge distillation. These components are used to find noise-resilient model parameters while accurately extracting knowledge from a potentially corrupted global model. Our client-side defense strategy has a flexible structure and can work in conjunction with any existing server-side strategies. Evaluations of real-world scenarios across multiple datasets show that the proposed method enhances the robustness of federated learning against model poisoning attacks.

27 January 2022 in IEEE Transactions on Knowledge and Data Engineering
Verified
Journal Article
Open access
SKSundong Kim
Sundong Kim
TMTung-Duong Mai
Tung-Duong Mai
SHSungwon Han
Sungwon Han
SPSungwon Park
Sungwon Park
DND.K. Thi Nguyen
D.K. Thi Nguyen
MCMeeyoung Cha
Meeyoung Cha
We study the human-in-the-loop customs inspection scenario, where an AI-assisted algorithm supports customs officers by recommending a set of imported goods to be inspected. If the inspected items are fraudulent, the officers can levy extra duties. These logs are then used as additional training data for the next iterations. Choosing to inspect suspicious items first leads to an immediate gain in customs revenue, yet such inspections may not bring new insights for learning dynamic traffic patterns. On the other hand, inspecting uncertain items can help acquire new knowledge, which will be used as a supplementary training resource to update the selection systems. Based on multiyear customs datasets from three countries, we demonstrate that some degree of exploration is necessary to cope with domain shifts in the trade data. The results show that a hybrid strategy of selecting likely fraudulent and uncertain items will eventually outperform the exploitation-only strategy.