Hyunsung Lee

Hyunsung Lee

Auto-generated profile, based on publicly available information

Hyunsung Lee

Department of Electrical and Computer Engineering, Sungkyunkwan University, Suwon, South Korea

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

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

Representative Publications

View All Publications
6 October 2020 in IEEE Access
Verified
Journal Article
Open access
HLHyunsung Lee
Hyunsung Lee
JLJinkyu Lee
Jinkyu Lee
IYIkjun Yeom
Ikjun Yeom
HWHonguk Woo
Honguk Woo
Recently, deep reinforcement learning (RL) technologies have been considered as a feasible solution for tackling combinatorial problems in various research and engineering areas. Motivated by this recent success of RL-based approaches, in this paper, we focus on how to utilize RL technologies in the context of real-time system research. Specifically, we first formulate the problem of fixed-priority assignments for multi-processor real-time scheduling, which has long been considered challenging in the real-time system community, as a combinatorial problem. We then propose the RL-based priority assignment model Panda that employs (i) a taskset embedding mechanism driven by attention-based encoder-decoder deep neural networks, hence enabling to efficiently extract useful features from the dynamic relation of periodic tasks. We also present two optimization schemes tailored to adopt RL for real-time task scheduling problems: (ii) the response time analysis (RTA)-based policy gradient RL and guided learning schemes, which facilitate the training processes of the Panda model. To the best of our knowledge, our approach is the first to employ RL for real-time task scheduling. Through various experiments, we show that Panda is competitive with well-known heuristic algorithms for real-time task scheduling upon a multi-processor platform, and it often outperforms them in large-scale non-trivial settings, e.g., achieving an average 7.7% enhancement in schedulability ratio for a testing system configuration of 64-sized tasksets and an 8-processor platform.

23 November 2021 in IEEE Access
Verified
Journal Article
Open access
HLHyunsung Lee
Hyunsung Lee
SCSangwoo Cho
Sangwoo Cho
YJYeongjae Jang
Yeongjae Jang
JLJinkyu Lee
Jinkyu Lee
HWHonguk Woo
Honguk Woo
Parallelization of tasks and efficient utilization of processors are considered important and challenging in operating large-scale real-time systems. Recently, deep reinforcement learning (DRL) was found to provide effective solutions to various combinatorial optimization problems. In this paper, inspired by recent achievements in DRL, we employ DRL techniques for scheduling a directed acyclic graph (DAG) task in which a set of non-preemptive subtasks are specified by precedence conditions among them. We propose a DRL-based priority assignment model for scheduling a DAG task on a multiprocessor system, named GoSu , which adapts a graph convolution network (GCN) to process a complex interdependent task structure and minimize the makespan of a DAG task. Our proposed model makes use of both temporal and structural features in a DAG to effectively learn a priority-based scheduling policy via GCN and policy gradient methods. With comprehensive evaluations, we verify that our model shows comparable performance to several state-of-the-art DAG task scheduling algorithms, and outperforms them by 2~3% in the slowdown of achieved makespans particularly in nontrivial system configurations where workloads are neither too small nor heavy compared to the given number of processors. We also analyze the priority assignment behaviors of our model by leveraging a regression method that imitates the learned policy of the model.

18 July 2023
Verified
Conference Paper
Open access
HLHyunsung Lee
Hyunsung Lee
SYSungwook Yoo
Sungwook Yoo
DLDongjun Lee
Dongjun Lee
JKJaekwang Kim
Jaekwang Kim
In real-world recommender model deployments, the models are typically retrained and deployed repeatedly. It is the rule-of-thumb to periodically retrain recommender models to capture up-to-date user behavior and item trends. However, the harm caused by delayed model updates has not been investigated extensively yet. in this perspective paper, we formulate the delayed model update problem and quantitatively demonstrate the delayed model update actually harms the model performance by increasing the number of cold users and cold items increase and decreasing overall model performances. These effects vary across different domains having different characteristics. Upon these findings, we further argue that although the delayed model update has negative effects on online recommender model deployment, yet it has not gathered enough attention from research communities. We argue our verification of the relationship between the model update cycle and model performance calls for further research such as faster model training, and more efficient data pipelines to keep the model more up-to-date with the latest user behaviors and item trends.

16 August 2021 in IEEE Access
Verified
Journal Article
Open access
HLHyunsung Lee
Hyunsung Lee
SCSangwoo Cho
Sangwoo Cho
YJYeongjae Jang
Yeongjae Jang
JKJaekwang Kim
Jaekwang Kim
HWHonguk Woo
Honguk Woo
Most recommenders generate recommendations for a user by computing the preference score of items, sorting the items according to the score, and filtering top- $K$ -items of high scores. Since sorting is not differentiable and is difficult to optimize with gradient descent, it is nontrivial to incorporate it in recommendation model training despite its relevance to top- $K$ recommendations. As a result, inconsistency occurs between existing learning objectives and ranking metrics of recommenders. In this work, we present the Differentiable Ranking Metric (DRM) that mitigates the inconsistency between model training and generating top- $K$ recommendations, aiming at improving recommendation performance by employing the differentiable relaxation of ranking metrics via joint learning. Using experiments with several real-world datasets, we demonstrate that the joint learning of the DRM objective and existing factor based recommenders significantly improves the quality of recommendations.

18 September 2022
Verified
Conference Paper
Open access
HLHyunsung Lee
Hyunsung Lee
SYSungwook Yoo
Sungwook Yoo
AYAndrew Yang
Andrew Yang
WJWon-Jun Jang
Won-Jun Jang
CPChiwan Park
Chiwan Park
In this paper, we describe our approach for the RecSys 2022 Challenge organized by Dressipi. The goal of the challenge is to predict which item is purchased next given sessions of users as well as metadata of items from fashion e-commerce service. One key characteristic of this problem is that most sessions only have few (lower than 3) previous views. Furthermore, a large number of sessions (about 19%) contain views and purchases of items that did not appear before. We propose the following approaches to overcome these problems. First, we introduce a simple, yet strong sequence-aware MLP that outperforms recently proposed sequential recommenders such as BERT4Rec and GRU4Rec in the given dataset. Secondly, we propose a similarity metric that captures not only item metadata but also item popularity. Lastly, we predict recommendations for different types of sessions with different serving models.