• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.12
• /
• pp.4776-4794
• /
• 2020

We propose a rapid adaptive learning framework for streaming object detection, called EER-ASSL. The method combines the expected error reduction (EER) dependent rollback learning and the active semi-supervised learning (ASSL) for a rapid adaptive CNN detector. Most CNN object detectors are built on the assumption of static data distribution. However, images are often noisy and biased, and the data distribution is imbalanced in a real world environment. The proposed method consists of collaborative sampling and EER-ASSL. The EER-ASSL utilizes the active learning (AL) and rollback based semi-supervised learning (SSL). The AL allows us to select more informative and representative samples measuring uncertainty and diversity. The SSL divides the selected streaming image samples into the bins and each bin repeatedly transfers the discriminative knowledge of the EER and CNN models to the next bin until convergence and incorporation with the EER rollback learning algorithm is achieved. The EER models provide a rapid short-term myopic adaptation and the CNN models an incremental long-term performance improvement. EER-ASSL can overcome noisy and biased labels in varying data distribution. Extensive experiments shows that EER-ASSL obtained 70.9 mAP compared to state-of-the-art technology such as Faster RCNN, SSD300, and YOLOv2.

• Journal of IKEEE
• /
• v.23 no.4
• /
• pp.1430-1439
• /
• 2019

In Functional Mockup Interface(FMI)-based co-simulation, a bisectional algorithm can be used to find the zerocrossing point as a way to improve the accuracy of the simulation results. In this paper, the proposed master algorithm(MA) analyzes the repeated interval graph and predicts the next interval by applying the Markov Chain to the step size. In the simulation, we propose an algorithm to minimize the rollback by storing the step size that changes according to the graph type as an array and applying it to the next prediction interval when the rollback occurs in the simulation. Simulation results show that the proposed algorithm reduces the simulation time by more than 20% compared to the existing algorithm.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.10 no.8
• /
• pp.3689-3700
• /
• 2016

It is not practically feasible to apply hardware-based fault-tolerant schemes, such as hardware replication, in mobile devices. Therefore, software-based fault-tolerance techniques, such as checkpoint and rollback schemes, are required. In checkpoint and rollback schemes, the optimal checkpoint interval should be applied to obtain the best performance. Most previous studies focused on minimizing the expected execution time or response time for completing a given task. Currently, most mobile applications run in real-time environments. Therefore, it is extremely essential for mobile devices to employ optimal checkpoint intervals as determined by the real-time constraints of tasks. In this study, we tackle the problem of determining the optimal inter-checkpoint interval of checkpoint and rollback schemes to maximize the deadline meet ratio in real-time systems and to build a probabilistic cost model. From this cost model, we can numerically find the optimal checkpoint interval using mathematical tools. The performance of the proposed solution is evaluated using analytical estimates.

• Proceedings of the Korea Society for Simulation Conference
• /
• 1999.10a
• /
• pp.331-336
• /
• 1999

웹은 지난 수년간 급속도로 발전하였으며 웹의 다양한 활용 분야 중에서 시뮬레이션은 웹의 특성을 가장 잘 이용하는 분야 중 하나로, 웹 기반 시뮬레이션의 구현에 대한 연구가 활발히 이루어지고 있다. 또한 Java 언어의 출현은 웹에서 실질적인 애니메이션과 애니메이션들간의 상호동작을 가능하게 한다. 웹 기반 분산 시뮬레이션은 웹의 분산 특성과 자바의 객체지향 특성을 이용한 분산 시뮬레이션이다. time-warp 기법을 사용하는 웹 기반 분산 시뮬레이션에서 speedup에 대한 성능은 rollback과 통신 지연이 가장 중요한 요인이다. rollback이 발생한 경우 시뮬레이션을 다시 수행하여 시뮬레이션을 매우 느리게 한다. 이러한 rollback과 통신 지연의 방대한 오버헤드는 시뮬레이션 모델의 지역적 분할을 사용할 때 발생한다. 본 발표에서는 time-warp을 기본 구졸 자바의 RMI를 사용하는 웹 기반 분산 시뮬레이션에서 통신 지연에 의한 오버헤드 및 거대한 병렬성과 분산을 고려한 시뮬레이션의 구현 모델을 제안하고 구현한다.

• Journal of KIISE
• /
• v.44 no.2
• /
• pp.132-138
• /
• 2017

We design and implement a process-based fault recovery system to increase the reliability of new memory based computer systems. A rollback point is made at every context switch to which a process can rollback to upon a fault. In this study, a clone process of the original process, which we refer to as a P-process (Persistent-process), is created as a rollback point. Such a design minimizes losses when a fault does occur. Specifically, first, execution loss can be minimized as rollback points are created only at context switches, which bounds the lost execution. Second, as we make use of the COW (Copy-On-Write）mechanism, only those parts of the process memory state that are modified (in page units) are copied decreasing the overhead for creating the P-process. Our experimental results show that the overhead is approximately 5% in 8 out of 11 PARSEC benchmark workloads when P-process is created at every context switch time. Even for workloads that result in considerable overhead, we show that this overhead can be reduced by increasing the P-process generation interval.

• Proceedings of the Korean Information Science Society Conference
• /
• 2007.06b
• /
• pp.513-518
• /
• 2007

시뮬레이션 방법론에 있어서 모델기반 시뮬레이션과 프로세스기반 시뮬레이션으로 나눌 수 있는데, 재사용성, 확장성, 시뮬레이터 기술 용이성 등의 장점으로 모델기반 시뮬레이션이 많이 사용되고 있다. 이러한 이유로 근래에는 컴퓨터 시스템, 항공, 자동차 등에서 모델 기반 시뮬레이션 방법이 사용되고 있다. 모델기반 시뮬레이션 방법으로 수학적 이론을 기반으로 모델을 정의하는 DEVS(Discrete Event System Specification) 형식론은 계층적이고 모듈화 된 형태로 이산사건 시스템을 기술한다. 대규모의 복잡한 시뮬레이션 모델을 검증 할 목적으로 분산 시뮬레이션 방법론이 있는데, 이들은 크게 동기적인 방법과 비동기적인 방법이 있다. 동기적 방식보다 빠른 수행을 위해 비동기적 방법은 전체 Time-order 순이 아닌 로컬 Time-order를 가진다. 그러나 비동기적 방식에는 분산된 시뮬레이터들 간의 전체 Time-order를 유지하기 위해 전 처리된 시뮬레이터 결과들을 저장하는데, Time-order 상으로 현재의 시뮬레이션 시간보다 과거의 사건이 왔을 때 그 이벤트를 처리해주어야 되기 때문이다. 이러한 비동기적 분산 시뮬레이션 방법론에서는 전체 Time-order를 유지하기 위해 과거의 Time-order를 가지는 이벤트가 왔을 때 rollback operation을 수행한다. 그러나 rollback operation은 분산 시뮬레이션 방법론에서 성능 장애요소 중 하나이다. 본 논문에서는 rollback operation을 최소할 할 수 있는 DEVS 모델 분배 방법을 제안한다.

• Journal of IKEEE
• /
• v.26 no.4
• /
• pp.628-632
• /
• 2022

In this paper, a triple-core delayed lock-step processor is proposed for automotive applications. It performs same operations in three different cores independently, and votes their results to get final values. Therefore its operations are safe even if errors occur in one core. Its three cores operate in a delayed manner to prevent simultaneous errors in multiple cores due to radiative ray or electromagnetic wave. When an error occurs in main core connected to the memory, wrong values can be stored in the memory, so the proposed processor performs memory rollback to restore correct values. Simulation results show that the proposed processor successfully compensates various errors.

• Journal of KIISE:Computer Systems and Theory
• /
• v.29 no.6
• /
• pp.331-339
• /
• 2002

This paper presents a garbage collection protocol for checkpoints and message logs which are staved on the stable storage or volatile storage for fault tolerancy. The previous works of garbage collections in coordinated checkpointing protocol delete all the checkpoints except for the last checkpoints on earth processes. But implemented in top of reliable communication protocol like as TCP/IP, rollback recovery protocol based on only last checkpoints makes sympathetic rollback. We show that the old checkpoints or message logs except for the last checkpoints have to be preserved in order to replay the lost message. And we define the conditions for garbage collection of checkpoints and message logs for lost messages and present the garbage collection algorithm for checkpoints and message logs in coordinated checkpointing protocol. Since the proposed algorithm uses process information for lost message piggybacked with messages, the additional messages for garbage collection is not required The proposed garbage collection algorithm makes 'the lazy garbage collectioneffect', because relying on the piggybacked checked checkpoint information in send/receive message. But 'the lazy garbage collection effect'does not break the consistency of the whole systems.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.14 no.3
• /
• pp.201-207
• /
• 2021

Deep learning, a type of machine learning, performs learning while changing the weights as it progresses through each learning process. Tensor Flow and Keras provide the results of the end of the learning in graph form. Thus, If an error occurs, the result must be discarded. Consequently, existing technologies provide a function to roll back learning results, but the rollback function is limited to results up to five times. Moreover, they applied the concept of MLOps to track the deep learning process, but no rollback capability is provided. In this paper, we construct a system that manages the intermediate value of the learning process by blockchain to record the intermediate learning process and can rollback in the event of an error. To perform the functions of blockchain, the deep learning process and the rollback of learning results are designed to work by writing Smart Contracts. Performance evaluation shows that, when evaluating the rollback function of the existing deep learning method, the proposed method has a 100% recovery rate, compared to the existing technique, which reduces the recovery rate after 6 times, down to 10% when 50 times. In addition, when using Smart Contract in Ethereum blockchain, it is confirmed that 1.57 million won is continuously consumed per block creation.

• ETRI Journal
• /
• v.4 no.4
• /
• pp.34-40
• /
• 1982