arXiv subjects

Hayoung Oh

Publications and source records attributed to Hayoung Oh.

6 recordsLinked to original sources

arXiv2026

A Multi-Prototype-Guided Federated Knowledge Distillation Approach in AI-RAN Enabled Multi-Access Edge Computing System

With the development of wireless network, Multi-Access Edge Computing (MEC) and Artificial Intelligence (AI)-native Radio Access Network (RAN) have attracted significant attention. Particularly, the integration of AI-RAN and MEC is envisioned to transform network efficiency and responsiveness. Therefore, it is valuable to investigate AI-RAN enabled MEC system. Federated learning (FL) nowadays is emerging as a promising approach for AI-RAN enabled MEC system, in which edge devices are enabled to train a global model cooperatively without revealing their raw data. However, conventional FL encounters the challenge in processing the non-independent and identically distributed (non-IID) data. Single prototype obtained by averaging the embedding vectors per class can be employed in FL to handle the data heterogeneity issue. Nevertheless, this may result in the loss of useful information owing to the average operation. Therefore, in this paper, a multi-prototype-guided federated knowledge distillation (MP-FedKD) approach is proposed. Particularly, self-knowledge distillation is integrated into FL to deal with the non-IID issue. To cope with the problem of information loss caused by single prototype-based strategy, multi-prototype strategy is adopted, where we present a conditional hierarchical agglomerative clustering (CHAC) approach and a prototype alignment scheme. Additionally, we design a novel loss function (called LEMGP loss) for each local client, where the relationship between global prototypes and local embedding will be focused. Extensive experiments over multiple datasets with various non-IID settings showcase that the proposed MP-FedKD approach outperforms the considered state-of-the-art baselines regarding accuracy, average accuracy and errors (RMSE and MAE).

cs.LG↗

arXiv2026

Geometric Knowledge-Assisted Federated Dual Knowledge Distillation Approach Towards Remote Sensing Satellite Imagery

Federated learning (FL) has recently become a promising solution for analyzing remote sensing satellite imagery (RSSI). However, the large scale and inherent data heterogeneity of images collected from multiple satellites, where the local data distribution of each satellite differs from the global one, present significant challenges to effective model training. To address this issue, we propose a Geometric Knowledge-Guided Federated Dual Knowledge Distillation (GK-FedDKD) framework for RSSI analysis. In our approach, each local client first distills a teacher encoder (TE) from multiple student encoders (SEs) trained with unlabeled augmented data. The TE is then connected with a shared classifier to form a teacher network (TN) that supervises the training of a new student network (SN). The intermediate representations of the TN are used to compute local covariance matrices, which are aggregated at the server to generate global geometric knowledge (GGK). This GGK is subsequently employed for local embedding augmentation to further guide SN training. We also design a novel loss function and a multi-prototype generation pipeline to stabilize the training process. Evaluation over multiple datasets showcases that the proposed GK-FedDKD approach is superior to the considered state-of-the-art baselines, e.g., the proposed approach with the Swin-T backbone surpasses previous SOTA approaches by an average 68.89% on the EuroSAT dataset.

cs.CV↗

arXiv2025

SNAP: A Plan-Driven Framework for Controllable Interactive Narrative Generation

Large Language Models (LLMs) hold great potential for web-based interactive applications, including browser games, online education, and digital storytelling platforms. However, LLM-based conversational agents suffer from spatiotemporal distortions when responding to variant user inputs, failing to maintain consistency with provided scenarios. We propose SNAP (Story and Narrative-based Agent with Planning), a framework that structures narratives into Cells with explicit Plans to prevent narrative drift in web environments. By confining context within each Cell and employing detailed plans that specify spatiotemporal settings, character actions, and plot developments, SNAP enables coherent and scenario-consistent dialogues while adapting to diverse user responses. Via automated and human evaluations, we validate SNAP's superiority in narrative controllability, demonstrating effective scenario consistency despite variant user inputs in web-based interactive storytelling.

cs.HC↗

arXiv2021

Improved Extrapolation Methods of Data-driven Background Estimation in High-Energy Physics

Data-driven methods of background estimations are often used to obtain more reliable descriptions of backgrounds. In hadron collider experiments, data-driven techniques are used to estimate backgrounds due to multi-jet events, which are difficult to model accurately. In this article, we propose an improvement on one of the most widely used data-driven methods in the hadron collision environment, the "ABCD" method of extrapolation. We describe the mathematical background behind the data-driven methods and extend the idea to propose improved general methods.

hep-ph↗

arXiv2020

Data-driven Estimation of Background Distribution through Neural Autoregressive Flows

We report on a general and automatic data-driven background distribution shape estimation method using neural autoregressive flows (NAF), which is one of the deep generative learning methods. Data-driven background estimation is indispensable for many analyses involving complicated final states where reliable predictions are not available. NAF allow us to construct general bijective transformations that operate on multidimensional space, out of finite number of invertible one-dimensional functions. Given its simplicity and universality, it is well suited to the application in the data-driven background estimation, since data-driven estimations can be expressed as transformations. In a data-driven background estimation, the goal is to derive appropriate transformations and apply extrapolated transformations to the region of interest. In the ABCDnn method, we can have the NAF learn the transformations' dependence on control variables by having multiple control regions. We demonstrate that the prediction through ABCDnn method is similar to optimal case, while having smaller statistical uncertainty.

hep-ph↗

arXiv2016

Follow Spam Detection based on Cascaded Social Information

In the last decade we have witnessed the explosive growth of online social networking services (SNSs) such as Facebook, Twitter, RenRen and LinkedIn. While SNSs provide diverse benefits for example, forstering interpersonal relationships, community formations and news propagation, they also attracted uninvited nuiance. Spammers abuse SNSs as vehicles to spread spams rapidly and widely. Spams, unsolicited or inappropriate messages, significantly impair the credibility and reliability of services. Therefore, detecting spammers has become an urgent and critical issue in SNSs. This paper deals with Follow spam in Twitter. Instead of spreading annoying messages to the public, a spammer follows (subscribes to) legitimate users, and followed a legitimate user. Based on the assumption that the online relationships of spammers are different from those of legitimate users, we proposed classification schemes that detect follow spammers. Particularly, we focused on cascaded social relations and devised two schemes, TSP-Filtering and SS-Filtering, each of which utilizes Triad Significance Profile (TSP) and Social status (SS) in a two-hop subnetwork centered at each other. We also propose an emsemble technique, Cascaded-Filtering, that combine both TSP and SS properties. Our experiments on real Twitter datasets demonstrated that the proposed three approaches are very practical. The proposed schemes are scalable because instead of analyzing the whole network, they inspect user-centered two hop social networks. Our performance study showed that proposed methods yield significantly better performance than prior scheme in terms of true positives and false positives.

cs.SI↗