In artificial intelligence (AI), many legal conflicts have arisen, especially concerning privacy and copyright associated with training data. When an AI model's training data incurs privacy concerns, it becomes imperative to develop a new model devoid of influences from such contentious data. However, retraining from scratch is often not viable due to the extensive data requirements and heavy computational costs. Machine unlearning presents a promising solution by enabling the selective erasure of specific knowledge from models. Despite its potential, many existing approaches in machine unlearning are based on scenarios that are either impractical or could lead to unintended degradation of model performance. We utilize the concept of weight prediction to approximate the less-learned weights based on observations about further training. By repetition of 1) finetuning on specific data and 2) weight prediction, our work gradually eliminates knowledge about the specific data. We verify its ability to eliminate side effects caused by problematic data and show its effectiveness across various architectures, datasets, and tasks.

Recent significant advancements in diffusion models have revolutionized image generation, enabling the synthesis of highly realistic images with text-based guidance. These breakthroughs have paved the way for constructing datasets via generative artificial intelligence (AI), offering immense potential for various applications. However, two critical challenges hinder the widespread adoption of synthesized data: computational cost and the generation of peculiar images. While computational costs have improved through various approaches, the issue of peculiar image generation remains relatively unexplored. Existing solutions rely on heuristics, extra training, or AI-based post-processing to mitigate this problem. In this paper, we present a novel approach to address both issues simultaneously. We establish that both gradient descent and diffusion sampling are specific cases of the generalized expectation maximization algorithm. We hypothesize and empirically demonstrate that peculiar image generation is akin to the local minima problem in optimization. Inspired by optimization techniques, we apply naive momentum and positive-negative momentum to diffusion sampling. Last, we propose new metrics to evaluate the peculiarity. Experimental results show momentum effectively prevents peculiar image generation without extra computation.

After the revelation that neural networks tend to produce overconfident predictions, the problem of calibration, which aims to align confidence with accuracy to enhance the reliability of predictions, has gained significant importance. Several solutions based on calibration maps have been proposed to address the problem of recalibrating a trained classifier using additional datasets. In this paper, we offer an algorithm that transforms the weights of the last layer of the classifier, distinct from the calibration-map-based approach. We concentrate on the geometry of the final linear layer, specifically its angular aspect, and adjust the weights of the corresponding layer. We name the method Tilt and Average, and validate the calibration effect empirically and theoretically. Through this, we demonstrate that our approach, in addition to the existing calibration-map-based techniques, can yield improved calibration performance.

The deep learning-based satellite image analysis and retraining systems are getting emerging technologies to enhance the capability of the sophisticated analysis of terrestrial objects. In principle, to apply the explainable DNN model for the process of satellite image analysis and retraining, we consider a new acceleration scheduling mechanism. Especially, the conventional DNN acceleration schemes cause serious performance degradation due to computational complexity and costs in satellite image analysis and retraining. In this article, to overcome the performance degradation, we propose cooperative scheduling schemes for explainable DNN acceleration in analysis and retraining process. For the purpose of it, we define the latency and energy cost modeling to derive the optimized processing time and cost required for explainable DNN acceleration. Especially, we show a minimum processing cost considered in the proposed scheduling via layer-level management of the explainable DNN on FPGA-GPU acceleration system. In addition, we evaluate the performance using an adaptive unlabeled data selection scheme with confidence threshold and a semi-supervised learning driven data parallelism scheme in accelerating retraining process. The experimental results demonstrate that the proposed schemes reduce the energy cost of the conventional DNN acceleration systems by up to about 40% while guaranteeing the latency constraints.

For smart grid services, accurate individual load forecasting is an essential element. When training individual forecasting models for multi-customers, discrepancies in data distribution among customers should be considered; there are two simple ways to build the models considering multi-customers: constructing each model independently or training as one model encompassing multi-customers. The independent approach shows higher accuracy than the latter. However, it deploys copious models, causing resource/management inefficiency; the latter is the opposite. A compromise between these two could be clustering-based forecasting. However, the previous studies are limited in applying to individual forecasting in that they focus on aggregated load and do not consider concept drift, which degrades accuracy over time. Therefore, we propose a distribution-aware temporal pooling framework that is enhanced clustering-based forecasting. For the clustering, we propose Variational Recurrent Deep Embedding (VaRDE) working in a distribution-aware manner, so it is suitable to process individual load. It allocates clusters to customers every time, so the clusters, where customers are assigned, are dynamically changed to resolve distribution change. We conducted experiments with real data for evaluation, and the result showed better performance than previous studies, especially with a few models even for unseen data, leading to high scalability.

As the deep learning (DL) has widely been used for application domains such as image classifications, natural language processing, and speech recognition, various software frameworks have been developed. They provide users with efficient programming interfaces for developing the DL applications. The optimization techniques within these frameworks generally are different from each other, which leads to different processing times for even the same applications. However, it is difficult that end users consider performance differences in processing time due to incompatible programming interface among the DL frameworks. These differences might cause redundant efforts and costs for end users to develop and maintain the applications. In this paper, we introduce an integrated deep learning engine (IDLE), a novel interface working on the top of the existing DL frameworks, which provides a convenient, flexible and scalable programming interface developing the DL applications for end users regardless of DL frameworks. Besides, we also propose a novel adaptive task scheduling scheme for training DL applications in a cluster with different GPUs. We implement our platform on the heterogeneous GPU cluster, and the results show that the proposed scheduling algorithm improves cost efficiency processing various DL applications.

To design the write buffer and flash translation layer (FTL) for a solid-state drive (SSD), previous studies have tried to increase overall SSD performance by parallel I/O and garbage collection overhead reduction. Recent works have proposed pattern-based managements, which uses the request size and read- or write-intensiveness to apply different policies to each type of data. In our observation, the locations of read and write requests are closely related, and the pattern of each type of data can be changed. In this work, we propose SUPA, a single unified read-write buffer and pattern-change-aware FTL on multi-channel SSD architecture. To increase both read and write hit ratios on the buffer based on locality, we use a single unified read-write buffer for both clean and dirty blocks. With proposed buffer, we can increase buffer hit ratio up to 8.0% and reduce 33.6% and 7.5% of read and write latencies, respectively. To handle pattern-changed blocks, we add a pattern handler between the buffer and the FTL, which monitors channel status and handles data by applying one of the two different policies according to the pattern changes. With pattern change handling process, we can reduce 1.0% and 15.4% of read and write latencies, respectively. In total, our evaluations show that SUPA can get up to 2.0 and 3.9 times less read and write latency, respectively, without loss of lifetime in comparison to previous works.

Multimedia transmissions for IoT (Internet-of-Things) sensing has a high demand of route capacity and tight requirements of end-to-end delay. In this paper, we address the problems on how to guarantee delay-related QoS requirements and to balance the energy consumption, while using multipath routing to offer high transmission capability for IoT multimedia sensing. This motivates us to design a Multipath Planning for Single-Source based transmissions routing scheme, namely MPSS, which establishes desirable multiple route paths following B-spline trajectories based on geographical information of source and sink node, sending and receiving angles, and inter-path distance. We further utilize a factor of hop distance to reduce the cumulated error of each hop due to the density of nodes, and to guarantee the delay-related QoS requirements. A Multipath Planning for Multi-Source routing scheme is also designed, namely MPMS, to assign the angle scope according to the source node's priority and traffic. Experimental results show that MPSS can effectively generate well-patterned multiple spline-based routes, and the end-to-end delay is under control according to the delay QoS requirement, while the total energy consumption is minimized.

As cloud computing and wireless body sensor network technologies become gradually developed, ubiquitous healthcare services prevent accidents instantly and effectively, as well as provides relevant information to reduce related processing time and cost. This study proposes a co-processing intermediary framework integrated cloud and wireless body sensor networks, which is mainly applied to fall detection and 3-D motion reconstruction. In this study, the main focuses includes distributed computing and resource allocation of processing sensing data over the computing architecture, network conditions and performance evaluation. Through this framework, the transmissions and computing time of sensing data are reduced to enhance overall performance for the services of fall events detection and 3-D motion reconstruction.

The development of an integrated and personalized healthcare system is becoming an important issue in the modern healthcare industry. One of main objectives of integrated healthcare system is to effectively manage patients having chronic disease. Different from acute disease, chronic disease requires long term care and its temporal information plays an important role to manage the status of disease. Thus, a patient having chronic disease needs to visit the hospital periodically, which generates large volume of medical data. Among the various chronic diseases, metabolic syndrome has become a major public healthcare issue in many countries. There have been efforts to develop a metabolic syndrome risk quantification and prediction model and to integrate them into personalized healthcare system, so as to predict the risk of having metabolic syndrome in the future. However, the development of methods for temporal progress management of metabolic syndrome has not been widely investigated. In this paper, we propose a method for identifying a temporal progress and patient's status of metabolic syndrome. Further, the effectiveness of the proposed method is evaluated using a sample patient data while emphasizing the capability to identify chronological changes of metabolic syndrome status.

Facing the increasing demands and challenges in the area of chronic disease care, various studies on the healthcare system which can, whenever and wherever, extract and process patient data have been conducted. Chronic diseases are the long-term diseases and require the processes of the real-time monitoring, multidimensional quantitative analysis, and the classification of patients' diagnostic information. A healthcare system for chronic diseases is characterized as an at-hospital and at-home service according to a targeted environment. Both services basically aim to provide patients with accurate diagnoses of disease by monitoring a variety of physical states with a number of monitoring methods, but there are differences between home and hospital environments, and the different characteristics should be considered in order to provide more accurate diagnoses for patients, especially, patients having chronic diseases. In this paper, we propose a patient status classification method for effectively identifying and classifying chronic diseases and show the validity of the proposed method. Furthermore, we present a new healthcare system architecture that integrates the at-home and at-hospital environment and discuss the applicability of the architecture using practical target services.

Electrical switching, when used to interconnect tens to hundreds of pods (each having a thousand of servers) in the core of a data center, incurs a high cost and power consumption and is expected to be replaced with optical switching soon. In this paper, we consider hypercube-based interconnection using optical switches in the core and study novel routing and wavelength assignment schemes for a new paradigm called Labeled Optical Burst Switching with Home Circuits (LOBS-HC). In particular, we propose a simple scheme called complementary HC assignment (CHA) for a 2-dimensional cube and ring, and extend the study to a n-cube (n >; 2) and generalized hypercube (GHC) by applying the concept of Spanning Balanced Trees (SBTs). We determine the number of wavelengths (and transceivers) needed in each case and show that it can be significantly lower than that needed with conventional wavelength routing using optical circuit switching (OCS). We also show compared the proposed solution with other proposed electronic or hybrid switching based solutions.

To make an immune-inspired network intrusion detection system (IDS) effective, this paper proposes a new framework, which includes our avidity-model based clonal selection (AMCS) algorithm as core element. The AMCS algorithm uses an improved representation for antigens (corresponding to network access patterns) and detectors (corresponding to detection rules). In particular, a bio-inspired technique called gene expression programming (GEP) is integrated with artificial immune system (AIS) in detector representation. In addition, inspired by the avidity model of immunology, this paper also defines new avidity/affinity functions (corresponding to the metric for quantify the interactions between detector and antigens) that take the priorities of attribute into account. Accordingly, the proposed algorithm integrates both negative selection and positive selection with a balance factor k to assign appropriate weights to self and non-self avidity. The well known KDD CUP'99 DATA set is used for performance evaluation. The results show that the intrusion detection based on AMCS provides a higher detection rate of DoS attack, a lower false alarm rate, and a lower detectors generation cost. Our results indicate that breaking the bottleneck of immune-inspired network IDS through adjusting basic elements is feasible and effective.

As computer and network technology grows, medical application is become more complex to solve the physiological problems within expected time. Workflow management systems (WMS) in grid computing are becoming more important to solve the sophisticated problem such as genomic analysis, drug discovery, disease identification, etc. Although existing WMS can provide basic management functionality in grid environment, consideration of user requirements such as performance, reliability and interaction with user is missing. In this paper, we discuss how to guarantee different user requirements according to user SLA in Grid workflow management system. A hybrid workflow management system for QoS constrained medical grid application is designed and implemented to provide QoS awareness WMS. The proposed system is applied to physio-grid e-health platform to identify human heart disease with ECG analysis and virtual heart simulation (VHS) workflow applications. The experimental result shows that the proposed system can flexibly enhance the physio-grid platform in terms of SLA guarantee of different user.

Heart disease is one of the most serious medical problems which threaten human life. Now there are several methods for detecting heart disease and ECG (electrocardiogram) signal analysis is one of the typical solutions. The other method such as MCG (magnetocardiogram) is also recommended for heart disease detection. In computer society, the e-Health systems using these diagnosis methods have been developed. However, each diagnosis method has their own weak points and also limitations in system performance according to the increase of data in quantity. Therefore, in order to improve each diagnosis method and conventional e-Health system, we propose and implement a novel integrated-diagnosis system combining grid technologies which is termed as a Physio-Grid system. We present the experimental and evaluation data in order to show our system capability in diagnosis and system performance. The results indicate that the proposed e-Health system can provide the plausible medical services and also it guarantees high reliability and system performance in data management and in diagnosis.

A grid technology is one of key issues for healthcare services provided in collaborative environments. In this paper, a SLA-adaptive workflow integrated grid resource management system for supporting collaborative heart disease simulator application in Physio-Grid service is proposed. At first, we propose the system architecture and framework that integrates workflow function into policy quorum based resource management (PQRM) system, one of existing grid resource management systems, for collaborative healthcare services. In addition, we derived the cost-adaptive policy adjustment scheme to adjust a gap between conditions and actions of workflow management and resource management policies for the proposed system. Based on this adjustment scheme, an appropriate policy can be selected according to QoS constraints of collaborated healthcare applications given by SLA negotiated with users. Finally, we evaluate proposed system using a collaborative heart disease simulation service and show the proposed system outperforms a general grid management system throughout performance comparisons under different types of SLA.

Picture archiving and communications system (PACS) is widely used in hospitals to take medical images, to send images to radiologists or doctors and to store images into storage in a digital manner, that has brought high efficiency in a medical process. Recently, PACS is required to evolve into a more sophisticated system that can communicate with other medical systems. The existing PACS was designed to be used in a Local Area Network (LAN) of a hospital and DICOM, the standard used in PACS, doesn't have any QoS for file transmission or security functions for an open network. Therefore, we propose a PACS system over Policy Quorum based Resource Management (PQRM) system, which manages storage resources and computing resources in Grid environment to provide satisfying advanced medical services with guaranteed QoS. Finally, we test the real situation medical scenario to prove the advanced functions our system provides and do several experiments to show the superiority of our novel system to existing PACS systems. The results indicate that while PQRM-based PACS providing the ability for advanced medical services it guarantees the transmission, provides best resource selection and disaster recovery capability.

In this paper, a local SNR aided sensor selecting (LSAS) algorithm is proposed to meet the energy efficiency requirement of wireless sensor networks (WSNs). In the conventional cluster routing protocols, during each round, all the subordinate sensors need to send their sensed information to their cluster heads, which is energy consuming. Realizing that it is not necessary to involve all the sensors due to the redundancy characteristic of the information on them, we propose that only those sensors with higher local SNR are selected to transmit their sensed data. Experimental results demonstrate that it is sufficient to only use the information on nodes with higher local SNR for target state estimation. The simulations also suggest that the proposed method consumes about 30%-50% less energy than the conventional method.

Peer-to-peer (P2P) technology benefits from a tractable and flexible job distribution scheme. However due to the dynamic nature of P2P, the optimal decision point of job scheduling is still unpredictable and some DLT-like heuristics are not sufficient to be a promising candidate for this role. As a reason of that, in this paper we have proposed the optimized time-constraint job distribution scheme in group based P2P networks. This scheme supports the hybrid type of jobs according to divisibility property. Also, it provides the methodology (3 phase-algorithms) which guarantees efficient job distribution under the condition of minimum completion time of executing jobs. Through the simulation in the aspects of balancing and minimum completion time, we will discuss about its performance evaluation.

Due to the dynamic nature of the grid computing and P2P in networks, the behavior of system might be very unpredictable in respect of reliability. Thus, we need well-organized system architecture to provide high system availability with resource scheduling scheme for P2P grid application. In this paper, we propose the SLA-constrained resource scheduling policy methodology in order to enhance system performance in P2P grid. The simulation results show that SCPA-based job scheduling can support the load balancing and guarantee the high system availability in system performance.

The delay-constrained heuristics for multicast routing algorithms capable of satisfying the quality of services requirements of real time applications are essential under distributed network environments. However, some of these heuristics may fail to provide a low cost tree as they assume that network links are symmetric. Furthermore, the time required constructing such a tree might be prohibitive, especially for large networks, as they employ a brute-force approach to search for low-cost delay-bounded paths among the route candidates. In this paper, we propose a new efficient algorithm considering delay variation constraints as well as with the total cost minimization scheme of the multicast trees. The extensive simulations show that the proposed algorithm satisfies the QoS requirement for real-time traffic, very short execution time and good scalability applicable to multicast groups in large networks.