ZNS SSDs (Zoned Namespace SSD) are block devices that provide stable performance and low price by forcing sequential writes, however their users have to pay the price to guarantee strong write order. In addition, to get the best performance from small zoned ZNS SSDs that give the users control over device-internal parallel elements, the users need to utilize the SSDs in detail. Due to these overheads, Ceph, one of the distributed storage systems, has up to 69% lower performance when using ZNS SSDs compared to using legacy SSD. In this paper, we present zCeph which solves the problems that occur when using small zoned ZNS SSD in storage systems. We implemented zCeph based on legacy Ceph and evaluated it using synthesized and real-world workloads, showing that the performance improved by up to 4.1x and 7x, respectively, compared to the legacy Ceph using ZNS SSD.
{"title":"zCeph: Achieving High Performance On Storage System Using Small Zoned ZNS SSD","authors":"Jin-Yong Ha, H. Yeom","doi":"10.1145/3555776.3577758","DOIUrl":"https://doi.org/10.1145/3555776.3577758","url":null,"abstract":"ZNS SSDs (Zoned Namespace SSD) are block devices that provide stable performance and low price by forcing sequential writes, however their users have to pay the price to guarantee strong write order. In addition, to get the best performance from small zoned ZNS SSDs that give the users control over device-internal parallel elements, the users need to utilize the SSDs in detail. Due to these overheads, Ceph, one of the distributed storage systems, has up to 69% lower performance when using ZNS SSDs compared to using legacy SSD. In this paper, we present zCeph which solves the problems that occur when using small zoned ZNS SSD in storage systems. We implemented zCeph based on legacy Ceph and evaluated it using synthesized and real-world workloads, showing that the performance improved by up to 4.1x and 7x, respectively, compared to the legacy Ceph using ZNS SSD.","PeriodicalId":42971,"journal":{"name":"Applied Computing Review","volume":null,"pages":null},"PeriodicalIF":1.0,"publicationDate":"2023-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89732334","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Edge servers must routinely undergo maintenance to ensure the environment's performance and security. During maintenance, applications hosted by outdated servers must be relocated to alternative servers to avoid downtime. In distributed edges with servers spread across large regions, ensuring that applications are not migrated to servers too far away from their users to avoid high latency hardens the maintenance planning. In addition, the limited power supply of edge sites restricts the list of suitable alternative hosts for the applications even further. Past work has focused on optimizing maintenance or increasing the power efficiency of edge computing infrastructures. However, no work addresses both objectives together. This paper presents Emma, a maintenance strategy that reduces power consumption during edge server maintenance without excessively extending maintenance time or increasing application latency. Experiments show that Emma can minimize power consumption during maintenance by up to 26.48% compared to strategies from the literature.
{"title":"Reducing Power Consumption during Server Maintenance on Edge Computing Infrastructures","authors":"Felipe Rubin, Paulo S. Souza, T. Ferreto","doi":"10.1145/3555776.3577739","DOIUrl":"https://doi.org/10.1145/3555776.3577739","url":null,"abstract":"Edge servers must routinely undergo maintenance to ensure the environment's performance and security. During maintenance, applications hosted by outdated servers must be relocated to alternative servers to avoid downtime. In distributed edges with servers spread across large regions, ensuring that applications are not migrated to servers too far away from their users to avoid high latency hardens the maintenance planning. In addition, the limited power supply of edge sites restricts the list of suitable alternative hosts for the applications even further. Past work has focused on optimizing maintenance or increasing the power efficiency of edge computing infrastructures. However, no work addresses both objectives together. This paper presents Emma, a maintenance strategy that reduces power consumption during edge server maintenance without excessively extending maintenance time or increasing application latency. Experiments show that Emma can minimize power consumption during maintenance by up to 26.48% compared to strategies from the literature.","PeriodicalId":42971,"journal":{"name":"Applied Computing Review","volume":null,"pages":null},"PeriodicalIF":1.0,"publicationDate":"2023-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89677045","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Over the past few years, blockchains have captured the public's interest with the promise of pseudo-anonymous decentralized exchange infrastructures. However, their potential is hindered by various technical issues, such as their ability to scale with problematic storage and communication costs and a fairly low transaction throughput. In this paper, we present SplitChain, a protocol intended to support the creation of scalable account-based blockchains without undermining decentralization and security. This is achieved by using sharding, i.e. by splitting the blockchain into several lighter chains managed by their own disjoint sets of validators called shards. These shards balance the load by processing disjoint sets of transactions in parallel. SplitChain distinguishes itself from other sharded blockchains by minimizing the synchronization constraints among shards while maintaining security guarantees. Finally, the protocol is designed to dynamically adapt the number of shards to the system load, to avoid over-dimensioning issues of most of the existing sharding-based solutions where the number of shards is static.
{"title":"Student Research Abstract: SplitChain: Blockchain with fully decentralized dynamic sharding resilient to fast adaptive adversaries","authors":"Arthur Rauch","doi":"10.1145/3555776.3577207","DOIUrl":"https://doi.org/10.1145/3555776.3577207","url":null,"abstract":"Over the past few years, blockchains have captured the public's interest with the promise of pseudo-anonymous decentralized exchange infrastructures. However, their potential is hindered by various technical issues, such as their ability to scale with problematic storage and communication costs and a fairly low transaction throughput. In this paper, we present SplitChain, a protocol intended to support the creation of scalable account-based blockchains without undermining decentralization and security. This is achieved by using sharding, i.e. by splitting the blockchain into several lighter chains managed by their own disjoint sets of validators called shards. These shards balance the load by processing disjoint sets of transactions in parallel. SplitChain distinguishes itself from other sharded blockchains by minimizing the synchronization constraints among shards while maintaining security guarantees. Finally, the protocol is designed to dynamically adapt the number of shards to the system load, to avoid over-dimensioning issues of most of the existing sharding-based solutions where the number of shards is static.","PeriodicalId":42971,"journal":{"name":"Applied Computing Review","volume":null,"pages":null},"PeriodicalIF":1.0,"publicationDate":"2023-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89892912","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Many privilege escalation exploits on Linux abuse vulnerable system calls to threaten the system's security. Therefore, various static and dynamic analysis based seccomp policy generation frameworks emerged. Yet, they either focus on a subset of the available binaries or are constrained by the inherent properties of dynamic, testing-based analysis, which are prone to false negatives. In this paper, we present Jesse, a static-analysis-based framework for generating seccomp policies for ELF binaries. We design and implement an abstract-interpretation-based constant propagation that helps the analyst identify vital system calls for arbitrary, non-obfuscated binaries. Using the extracted results, Jesse allows producing effective seccomp policies, reducing the system's attack vector. To assess Jesse's effectiveness and accuracy, we have applied our system to over 1,000 ELF binaries for Debian 10, and show that---contrary to existing solutions---Jesse produces accurate and safely approximated results, without relying on any properties of the target binaries. In addition, we conduct a case study in which we combine Jesse's constant propagation strategy with container debloating techniques to produce seccomp policies that restrict up to five times more system calls than the Docker's default seccomp policy on average.
{"title":"Free Willy: Prune System Calls to Enhance Software Security","authors":"Charlie Groh, Sergej Proskurin, Apostolis Zarras","doi":"10.1145/3555776.3577593","DOIUrl":"https://doi.org/10.1145/3555776.3577593","url":null,"abstract":"Many privilege escalation exploits on Linux abuse vulnerable system calls to threaten the system's security. Therefore, various static and dynamic analysis based seccomp policy generation frameworks emerged. Yet, they either focus on a subset of the available binaries or are constrained by the inherent properties of dynamic, testing-based analysis, which are prone to false negatives. In this paper, we present Jesse, a static-analysis-based framework for generating seccomp policies for ELF binaries. We design and implement an abstract-interpretation-based constant propagation that helps the analyst identify vital system calls for arbitrary, non-obfuscated binaries. Using the extracted results, Jesse allows producing effective seccomp policies, reducing the system's attack vector. To assess Jesse's effectiveness and accuracy, we have applied our system to over 1,000 ELF binaries for Debian 10, and show that---contrary to existing solutions---Jesse produces accurate and safely approximated results, without relying on any properties of the target binaries. In addition, we conduct a case study in which we combine Jesse's constant propagation strategy with container debloating techniques to produce seccomp policies that restrict up to five times more system calls than the Docker's default seccomp policy on average.","PeriodicalId":42971,"journal":{"name":"Applied Computing Review","volume":null,"pages":null},"PeriodicalIF":1.0,"publicationDate":"2023-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90065825","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Building conversational agents to help humans in domain-specific tasks is challenging since the agent needs to understand the natural language and act over it while accessing domain expert knowledge. Modern natural language processing techniques led to an expansion of conversational agents, with recent pretrained language models achieving increasingly accurate language recognition results using ever-larger open datasets. However, the black-box nature of such pretrained language models obscures the agent's reasoning and its motivations when responding, leading to unexplained dialogues. We develop a belief-desire-intention (BDI) agent as a task-oriented dialogue system to introduce mental attitudes similar to humans describing their behavior during a dialogue. We compare the resulting model with a pipeline dialogue model by leveraging existing components from dialogue systems and developing the agent's intention selection as a dialogue policy. We show that combining traditional agent modelling approaches, such as BDI, with more recent learning techniques can result in efficient and scrutable dialogue systems.
{"title":"Modeling a Conversational Agent using BDI Framework","authors":"Alexandre Yukio Ichida, Felipe Meneguzzi","doi":"10.1145/3555776.3577657","DOIUrl":"https://doi.org/10.1145/3555776.3577657","url":null,"abstract":"Building conversational agents to help humans in domain-specific tasks is challenging since the agent needs to understand the natural language and act over it while accessing domain expert knowledge. Modern natural language processing techniques led to an expansion of conversational agents, with recent pretrained language models achieving increasingly accurate language recognition results using ever-larger open datasets. However, the black-box nature of such pretrained language models obscures the agent's reasoning and its motivations when responding, leading to unexplained dialogues. We develop a belief-desire-intention (BDI) agent as a task-oriented dialogue system to introduce mental attitudes similar to humans describing their behavior during a dialogue. We compare the resulting model with a pipeline dialogue model by leveraging existing components from dialogue systems and developing the agent's intention selection as a dialogue policy. We show that combining traditional agent modelling approaches, such as BDI, with more recent learning techniques can result in efficient and scrutable dialogue systems.","PeriodicalId":42971,"journal":{"name":"Applied Computing Review","volume":null,"pages":null},"PeriodicalIF":1.0,"publicationDate":"2023-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74685130","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
L. Bongiovanni, Luca Bruno, Fabrizio Dominici, Giuseppe Rizzo
Classification of documents according to a custom internal hierarchical taxonomy is a common problem for many organizations that deal with textual data. Approaches aimed to address this challenge are, for the vast majority, supervised methods, which have the advantage of producing good results on specific datasets, but the major drawbacks of requiring an entire corpus of annotated documents, and the resulting models are not directly applicable to a different taxonomy. In this paper, we aim to contribute to this important issue, by proposing a method to classify text according to a custom hierarchical taxonomy entirely without the need of labelled data. The idea is to first leverage the semantic information encoded into pre-trained Deep Language Models to assigned a prior relevance score for each label of the taxonomy using zero-shot, and secondly take advantage of the hierarchical structure to reinforce this prior belief. Experiments are conducted on three hierarchically annotated datasets: WebOfScience, DBpedia Extracts and Amazon Product Reviews, which are very diverse in the type of language adopted and have taxonomy depth of two and three levels. We first compare different zero-shot methods, and then we show that our hierarchy-aware approach substantially improves results across every dataset.
{"title":"Zero-Shot Taxonomy Mapping for Document Classification","authors":"L. Bongiovanni, Luca Bruno, Fabrizio Dominici, Giuseppe Rizzo","doi":"10.1145/3555776.3577653","DOIUrl":"https://doi.org/10.1145/3555776.3577653","url":null,"abstract":"Classification of documents according to a custom internal hierarchical taxonomy is a common problem for many organizations that deal with textual data. Approaches aimed to address this challenge are, for the vast majority, supervised methods, which have the advantage of producing good results on specific datasets, but the major drawbacks of requiring an entire corpus of annotated documents, and the resulting models are not directly applicable to a different taxonomy. In this paper, we aim to contribute to this important issue, by proposing a method to classify text according to a custom hierarchical taxonomy entirely without the need of labelled data. The idea is to first leverage the semantic information encoded into pre-trained Deep Language Models to assigned a prior relevance score for each label of the taxonomy using zero-shot, and secondly take advantage of the hierarchical structure to reinforce this prior belief. Experiments are conducted on three hierarchically annotated datasets: WebOfScience, DBpedia Extracts and Amazon Product Reviews, which are very diverse in the type of language adopted and have taxonomy depth of two and three levels. We first compare different zero-shot methods, and then we show that our hierarchy-aware approach substantially improves results across every dataset.","PeriodicalId":42971,"journal":{"name":"Applied Computing Review","volume":null,"pages":null},"PeriodicalIF":1.0,"publicationDate":"2023-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79385181","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Hugo Sousa, Arian Pasquali, Alípio Jorge, Catarina Sousa Santos, M'ario Amorim Lopes
Textual health records of cancer patients are usually protracted and highly unstructured, making it very time-consuming for health professionals to get a complete overview of the patient's therapeutic course. As such limitations can lead to suboptimal and/or inefficient treatment procedures, healthcare providers would greatly benefit from a system that effectively summarizes the information of those records. With the advent of deep neural models, this objective has been partially attained for English clinical texts, however, the research community still lacks an effective solution for languages with limited resources. In this paper, we present the approach we developed to extract procedures, drugs, and diseases from oncology health records written in European Portuguese. This project was conducted in collaboration with the Portuguese Institute for Oncology which, besides holding over 10 years of duly protected medical records, also provided oncologist expertise throughout the development of the project. Since there is no annotated corpus for biomedical entity extraction in Portuguese, we also present the strategy we followed in annotating the corpus for the development of the models. The final models, which combined a neural architecture with entity linking, achieved F1 scores of 88.6, 95.0, and 55.8 per cent in the mention extraction of procedures, drugs, and diseases, respectively.
{"title":"A Biomedical Entity Extraction Pipeline for Oncology Health Records in Portuguese","authors":"Hugo Sousa, Arian Pasquali, Alípio Jorge, Catarina Sousa Santos, M'ario Amorim Lopes","doi":"10.1145/3555776.3578577","DOIUrl":"https://doi.org/10.1145/3555776.3578577","url":null,"abstract":"Textual health records of cancer patients are usually protracted and highly unstructured, making it very time-consuming for health professionals to get a complete overview of the patient's therapeutic course. As such limitations can lead to suboptimal and/or inefficient treatment procedures, healthcare providers would greatly benefit from a system that effectively summarizes the information of those records. With the advent of deep neural models, this objective has been partially attained for English clinical texts, however, the research community still lacks an effective solution for languages with limited resources. In this paper, we present the approach we developed to extract procedures, drugs, and diseases from oncology health records written in European Portuguese. This project was conducted in collaboration with the Portuguese Institute for Oncology which, besides holding over 10 years of duly protected medical records, also provided oncologist expertise throughout the development of the project. Since there is no annotated corpus for biomedical entity extraction in Portuguese, we also present the strategy we followed in annotating the corpus for the development of the models. The final models, which combined a neural architecture with entity linking, achieved F1 scores of 88.6, 95.0, and 55.8 per cent in the mention extraction of procedures, drugs, and diseases, respectively.","PeriodicalId":42971,"journal":{"name":"Applied Computing Review","volume":null,"pages":null},"PeriodicalIF":1.0,"publicationDate":"2023-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"75786805","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Hind Bangui, Emilia Cioroaica, Mouzhi Ge, Barbora Buhnova
Internet of Behavior (IoB) has emerged as a new research paradigm within the context of digital ecosystems, with the support for understanding and positively influencing human behavior by merging behavioral sciences with information technology, and fostering mutual trust building between humans and technology. For example, when automated systems identify improper human driving behavior, IoB can support integrated behavioral adaptation to avoid driving risks that could lead to hazardous situations. In this paper, we propose an ecosystem-level self-adaptation mechanism that aims to provide runtime evidence for trust building in interaction among IoB elements. Our approach employs an indirect trust management scheme based on deep learning, which has the ability to mimic human behaviour and trust building patterns. In order to validate the model, we consider Pay-How-You-Drive vehicle insurance as a showcase of a IoB application aiming to advance the adaptation of business incentives based on improving driver behavior profiling. The experimental results show that the proposed model can identify different driving states with high accuracy, to support the IoB applications.
行为互联网(Internet of Behavior, IoB)是数字生态系统背景下的一种新的研究范式，通过将行为科学与信息技术相结合，促进人与技术之间的相互信任，支持理解和积极影响人类行为。例如，当自动系统识别出人类不当驾驶行为时，IoB可以支持综合行为适应，以避免可能导致危险情况的驾驶风险。在本文中，我们提出了一种生态系统级的自适应机制，旨在为IoB元素之间相互作用中的信任建立提供运行时证据。我们的方法采用了基于深度学习的间接信任管理方案，该方案具有模仿人类行为和信任建立模式的能力。为了验证该模型，我们将按需付费汽车保险作为IoB应用的一个展示，该应用旨在通过改进驾驶员行为分析来促进商业激励的适应。实验结果表明，该模型能较准确地识别不同的驾驶状态，支持IoB应用。
{"title":"Deep-Learning based Trust Management with Self-Adaptation in the Internet of Behavior","authors":"Hind Bangui, Emilia Cioroaica, Mouzhi Ge, Barbora Buhnova","doi":"10.1145/3555776.3577694","DOIUrl":"https://doi.org/10.1145/3555776.3577694","url":null,"abstract":"Internet of Behavior (IoB) has emerged as a new research paradigm within the context of digital ecosystems, with the support for understanding and positively influencing human behavior by merging behavioral sciences with information technology, and fostering mutual trust building between humans and technology. For example, when automated systems identify improper human driving behavior, IoB can support integrated behavioral adaptation to avoid driving risks that could lead to hazardous situations. In this paper, we propose an ecosystem-level self-adaptation mechanism that aims to provide runtime evidence for trust building in interaction among IoB elements. Our approach employs an indirect trust management scheme based on deep learning, which has the ability to mimic human behaviour and trust building patterns. In order to validate the model, we consider Pay-How-You-Drive vehicle insurance as a showcase of a IoB application aiming to advance the adaptation of business incentives based on improving driver behavior profiling. The experimental results show that the proposed model can identify different driving states with high accuracy, to support the IoB applications.","PeriodicalId":42971,"journal":{"name":"Applied Computing Review","volume":null,"pages":null},"PeriodicalIF":1.0,"publicationDate":"2023-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"80312584","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Dynamic Time Warping (DTW) is a robust method to measure the similarity between two sequences. This paper proposes a method based on DTW to analyse high-speed data streams. The central idea is to decompose the network traffic into sequences of histograms of packet sizes and then calculate the distance between pairs of such sequences using DTW with Kullback-Leibler (KL) distance. As a baseline, we also compute the Euclidean Distance between the sequences of histograms. Since our preliminary experiments indicate that the distance between two sequences falls within a different range of values for distinct types of streams, we then exploit this distance information for stream classification using a Random Forest. The approach was investigated using recent internet traffic data from a telecommunications company. To illustrate the application of our approach, we conducted a case study with encrypted Internet Protocol Television (IPTV) network traffic data. The goal was to use our DTW-based approach to detect the video codec used in the streams, as well as the IPTV channel. Results strongly suggest that the DTW distance value between the data streams is highly informative for such classification tasks.
{"title":"A DTW Approach for Complex Data A Case Study with Network Data Streams","authors":"Paula Raissa Silva, João Vinagre, J. Gama","doi":"10.1145/3555776.3577638","DOIUrl":"https://doi.org/10.1145/3555776.3577638","url":null,"abstract":"Dynamic Time Warping (DTW) is a robust method to measure the similarity between two sequences. This paper proposes a method based on DTW to analyse high-speed data streams. The central idea is to decompose the network traffic into sequences of histograms of packet sizes and then calculate the distance between pairs of such sequences using DTW with Kullback-Leibler (KL) distance. As a baseline, we also compute the Euclidean Distance between the sequences of histograms. Since our preliminary experiments indicate that the distance between two sequences falls within a different range of values for distinct types of streams, we then exploit this distance information for stream classification using a Random Forest. The approach was investigated using recent internet traffic data from a telecommunications company. To illustrate the application of our approach, we conducted a case study with encrypted Internet Protocol Television (IPTV) network traffic data. The goal was to use our DTW-based approach to detect the video codec used in the streams, as well as the IPTV channel. Results strongly suggest that the DTW distance value between the data streams is highly informative for such classification tasks.","PeriodicalId":42971,"journal":{"name":"Applied Computing Review","volume":null,"pages":null},"PeriodicalIF":1.0,"publicationDate":"2023-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"77968707","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Combining the challenges of streaming data and multi-label learning, the task of mining a drifting, multi-label data stream requires methods that can accurately predict labelsets, adapt to various types of concept drift and run fast enough to process each data point before the next arrives. To achieve greater accuracy, many multi-label algorithms use computationally expensive techniques, such as multiple adaptive windows, with little concern for runtime and memory complexity. We present Aging and Rejuvenating kNN (ARkNN) which uses simple resources and efficient strategies to weight instances based on age, predictive performance, and similarity to the incoming data. We break down ARkNN into its component strategies to show the impact of each and experimentally compare ARkNN to seven state-of-the-art methods for learning from multi-label data streams. We demonstrate that it is possible to achieve competitive performance in multi-label classification on streams without sacrificing runtime and memory use, and without using complex and computationally expensive dual memory strategies.
{"title":"Aging and rejuvenating strategies for fading windows in multi-label classification on data streams","authors":"M. Roseberry, S. Džeroski, A. Bifet, Alberto Cano","doi":"10.1145/3555776.3577625","DOIUrl":"https://doi.org/10.1145/3555776.3577625","url":null,"abstract":"Combining the challenges of streaming data and multi-label learning, the task of mining a drifting, multi-label data stream requires methods that can accurately predict labelsets, adapt to various types of concept drift and run fast enough to process each data point before the next arrives. To achieve greater accuracy, many multi-label algorithms use computationally expensive techniques, such as multiple adaptive windows, with little concern for runtime and memory complexity. We present Aging and Rejuvenating kNN (ARkNN) which uses simple resources and efficient strategies to weight instances based on age, predictive performance, and similarity to the incoming data. We break down ARkNN into its component strategies to show the impact of each and experimentally compare ARkNN to seven state-of-the-art methods for learning from multi-label data streams. We demonstrate that it is possible to achieve competitive performance in multi-label classification on streams without sacrificing runtime and memory use, and without using complex and computationally expensive dual memory strategies.","PeriodicalId":42971,"journal":{"name":"Applied Computing Review","volume":null,"pages":null},"PeriodicalIF":1.0,"publicationDate":"2023-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"76923236","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：