• Journal of Korea Game Society
• /
• v.19 no.3
• /
• pp.5-14
• /
• 2019

The gaming server is based on a distributed server. In order to distribute workloads of gaming servers, distributed gaming servers apply some algorithms which divide each of gaming server's workload into balanced workload among the gaming servers and as a result, efficiently manage response time and fusibility of server requested by the clients. In this paper, we propose a load balancing agent using PPO(Proximal Policy Optimization) which is one of the methods from a greedy algorithm and Policy Gradient which is from Reinforcement Learning. The proposed load balancing agent is compared with the previous researches based on the simulation.

• Journal of Information Processing Systems
• /
• v.17 no.2
• /
• pp.297-305
• /
• 2021

Large amount of data is being generated in gaming servers due to the increase in the number of users and the variety of game services being provided. In particular, load balancing schemes for gaming servers are crucial consideration. The existing literature proposes algorithms that distribute loads in servers by mostly concentrating on load balancing and cooperative offloading. However, many proposed schemes impose heavy restrictions and assumptions, and such a limited service classification method is not enough to satisfy the wide range of service requirements. We propose a load balancing agent that combines the dynamic allocation programming method, a type of greedy algorithm, and proximal policy optimization, a reinforcement learning. Also, we compare performances of our proposed scheme and those of a scheme from previous literature, ProGreGA, by running a simulation.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.6
• /
• pp.1877-1891
• /
• 2022

As a part of cloud computing technology, algorithms for cloud task scheduling place an important influence on the area of cloud computing in data centers. In our earlier work, we proposed DeepEnergyJS, which was designed based on the original version of the policy gradient and reinforcement learning algorithm. We verified its effectiveness through simulation experiments. In this study, we used the Proximal Policy Optimization (PPO) algorithm to update DeepEnergyJS to DeepEnergyJSV2.0. First, we verify the convergence of the PPO algorithm on the dataset of Alibaba Cluster Data V2018. Then we contrast it with reinforcement learning algorithm in terms of convergence rate, converged value, and stability. The results indicate that PPO performed better in training and test data sets compared with reinforcement learning algorithm, as well as other general heuristic algorithms, such as First Fit, Random, and Tetris. DeepEnergyJSV2.0 achieves better energy efficiency than DeepEnergyJS by about 7.814%.

• Journal of the Korea Society for Simulation
• /
• v.27 no.4
• /
• pp.1-8
• /
• 2018

The most common way to control a target point using artificial intelligence is through reinforcement learning. However, it had to process complicated calculations that were difficult to implement in order to process reinforcement learning. In this paper, the enhanced Proximal Policy Optimization (PPO) algorithm was used to simulate finding the planned flight trajectory to reach the target point in the virtual environment. In this paper, we simulated how this problem was used to find the planned flight trajectory to reach the target point in the virtual environment using the enhanced Proximal Policy Optimization(PPO) algorithm. In addition, variables such as changes in trajectory, effects of rewards, and external winds are added to determine the zero conditions of external environmental factors on flight trajectory learning, and the effects on trajectory learning performance and learning speed are compared. From this result, the simulation results have shown that the agent can find the optimal trajectory in spite of changes in the various external environments, which will be applicable to the actual vehicle.

• Journal of Korea Game Society
• /
• v.19 no.6
• /
• pp.61-70
• /
• 2019

In this paper, we apply reinforcement learning in the field of management simulation game to check whether game agents achieve autonomously given goal. In this system, we apply PPO (Proximal Policy Optimization) algorithm in the Unity Machine Learning (ML) Agent environment and the game agent is designed to automatically find a way to play. Five game scenario simulation experiments were conducted to verify their usefulness. As a result, it was confirmed that the game agent achieves the goal through learning despite the change of environment variables in the game.

• Journal of the Korean Society of Industry Convergence
• /
• v.25 no.4_2
• /
• pp.645-653
• /
• 2022

Recently, deep reinforcement learning has been applied to a variety of industries, such as games, robotics, autonomous vehicles, and data cooling systems. An algorithm called reinforcement learning allows for automated asset allocation without the requirement for ongoing monitoring. It is free to choose its own policies. The purpose of this paper is to carry out an empirical analysis of the performance of asset allocation strategies. Among the strategies considered were the conventional Mean- Variance Optimization (MVO) and the Proximal Policy Optimization (PPO). According to the findings, the PPO outperformed both its benchmark index and the MVO. This paper demonstrates how dynamic asset allocation can benefit from the development of a reinforcement learning algorithm.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.2
• /
• pp.61-66
• /
• 2024

This paper aims to design a simulation for path-finding of objects in a simulation or game environment using reinforcement learning techniques. The main feature of this study is that the objects in the simulation are trained to avoid obstacles at random locations generated on a given track and to automatically explore path to get items. To implement the simulation, ML Agents provided by Unity Game Engine were used, and a learning policy based on PPO (Proximal Policy Optimization) was established to form a reinforcement learning environment. Through the reinforcement learning-based simulation designed in this study, we were able to confirm that the object moves on the track by avoiding obstacles and exploring path to acquire items as it learns, by analyzing the simulation results and learning result graph.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.5
• /
• pp.179-186
• /
• 2021

Manipulation of complex objects with an anthropomorphic robot hand like a human hand is a challenge in the human-centric environment. In order to train the anthropomorphic robot hand which has a high degree of freedom (DoF), human demonstration augmented deep reinforcement learning policy optimization methods have been proposed. In this work, we first demonstrate augmentation of human demonstration in deep reinforcement learning (DRL) is effective for object manipulation by comparing the performance of the augmentation-free Natural Policy Gradient (NPG) and Demonstration Augmented NPG (DA-NPG). Then three DRL policy optimization methods, namely NPG, Trust Region Policy Optimization (TRPO), and Proximal Policy Optimization (PPO), have been evaluated with DA (i.e., DA-NPG, DA-TRPO, and DA-PPO) and without DA by manipulating six objects such as apple, banana, bottle, light bulb, camera, and hammer. The results show that DA-NPG achieved the average success rate of 99.33% whereas NPG only achieved 60%. In addition, DA-NPG succeeded grasping all six objects while DA-TRPO and DA-PPO failed to grasp some objects and showed unstable performances.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.11
• /
• pp.4268-4289
• /
• 2020

Advanced traffic signal timing method plays very important role in reducing road congestion and air pollution. Reinforcement learning is considered as superior approach to build traffic light timing scheme by many recent studies. It fulfills real adaptive control by the means of taking real-time traffic information as state, and adjusting traffic light scheme as action. However, existing works behave inefficient in complex intersections and they are lack of feasibility because most of them adopt traffic light scheme whose phase sequence is flexible. To address these issues, a novel adaptive traffic signal timing scheme is proposed. It's based on actor-critic reinforcement learning algorithm, and advanced techniques proximal policy optimization and generalized advantage estimation are integrated. In particular, a new kind of reward function and a simplified form of state representation are carefully defined, and they facilitate to improve the learning efficiency and reduce the computational complexity, respectively. Meanwhile, a fixed phase sequence signal scheme is derived, and constraint on the variations of successive phase durations is introduced, which enhances its feasibility and robustness in field applications. The proposed scheme is verified through field-data-based experiments in both medium and high traffic density scenarios. Simulation results exhibit remarkable improvement in traffic performance as well as the learning efficiency comparing with the existing reinforcement learning-based methods such as 3DQN and DDQN.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.05a
• /
• pp.446-447
• /
• 2022

대부분의 머신러닝 및 딥러닝 모델의 경우 하이퍼 파라미터 선택은 모델의 성능에 큰 영향을 미친다. 따라서 전문가들은 작업을 수행하기 위해 모델을 구축할 때 하이퍼 파라미터 튜닝을 수행하는 데 상당한 시간을 소비해야 한다. Hyperparameter Optimization(HPO)을 해결하기 위한 알고리즘은 많지만 대부분의 방법은 검색을 수행하기 위해 각 epoch에서 실제 실험 결과를 필요로 한다. 따라서 HPO 검색을 위한 시간과 계산 지원을 줄이기 위해 본 논문에서는 Multi-agent Proximal Policy Optimization(MAPPO) 강화 학습 알고리즘을 제안한다. 2개의 이미지 분류 데이터 세트에 대한 실험 결과는 우리의 모델이 속도와 정확성에서 다른 기존 방법보다 우수하다는 것을 보여준다.