Due to the pervasiveness of group activities in people’s daily life, group recommendation has attracted a massive research effort in both industry and academia. A fundamental challenge in group recommendation is how to aggregate the preferences of group members to select a set of items maximizing the overall satisfaction of the group; this is the focus of this paper. Specifically, we introduce a dual adjustment aggregation score, which measures the relevance of an item to a group. We then propose a recommendation scheme, termed 𝑘 -dual adjustment unanimous skyline, that seeks to retrieve the 𝑘 items with the highest score, while discarding items that are unanimously considered inap-propriate. Furthermore, we design and develop algorithms for computing the 𝑘 -dual adjustment unanimous skyline efficiently. Finally, we demonstrate both the retrieval effectiveness and the efficiency of our approach through an extensive experimental evaluation on real datasets.
{"title":"Recommending Unanimously Preferred Items to Groups","authors":"Karim Benouaret, K. Tan","doi":"10.48786/edbt.2023.29","DOIUrl":"https://doi.org/10.48786/edbt.2023.29","url":null,"abstract":"Due to the pervasiveness of group activities in people’s daily life, group recommendation has attracted a massive research effort in both industry and academia. A fundamental challenge in group recommendation is how to aggregate the preferences of group members to select a set of items maximizing the overall satisfaction of the group; this is the focus of this paper. Specifically, we introduce a dual adjustment aggregation score, which measures the relevance of an item to a group. We then propose a recommendation scheme, termed 𝑘 -dual adjustment unanimous skyline, that seeks to retrieve the 𝑘 items with the highest score, while discarding items that are unanimously considered inap-propriate. Furthermore, we design and develop algorithms for computing the 𝑘 -dual adjustment unanimous skyline efficiently. Finally, we demonstrate both the retrieval effectiveness and the efficiency of our approach through an extensive experimental evaluation on real datasets.","PeriodicalId":88813,"journal":{"name":"Advances in database technology : proceedings. International Conference on Extending Database Technology","volume":"116 1","pages":"364-377"},"PeriodicalIF":0.0,"publicationDate":"2023-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89386677","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}
Data narration is the process of telling stories with insights ex-tracted from data. It is an instance of data science [4] where the pipeline focuses on data collection and exploration, answering questions, structuring answers, and finally presenting them to stakeholders [16, 17]. This tutorial reviews the challenges and opportunities of the full and semi-automation of these steps. In doing so, it draws from the extensive literature in data narration, data exploration and data visualization. In particular, we point out key theoretical and practical contributions in each domain such as next-step recommendation and policy learning for data exploration, insight interestingness and evaluation frameworks, and the crafting of data stories for the people who will exploit them. We also identify topics that are still worth investigating, such as the inclusion of different stakeholders’ profiles in designing data pipelines with the goal of providing data narration for all.
{"title":"Data Narration for the People: Challenges and Opportunities","authors":"S. Amer-Yahia, Patrick Marcel, Verónika Peralta","doi":"10.48786/edbt.2023.82","DOIUrl":"https://doi.org/10.48786/edbt.2023.82","url":null,"abstract":"Data narration is the process of telling stories with insights ex-tracted from data. It is an instance of data science [4] where the pipeline focuses on data collection and exploration, answering questions, structuring answers, and finally presenting them to stakeholders [16, 17]. This tutorial reviews the challenges and opportunities of the full and semi-automation of these steps. In doing so, it draws from the extensive literature in data narration, data exploration and data visualization. In particular, we point out key theoretical and practical contributions in each domain such as next-step recommendation and policy learning for data exploration, insight interestingness and evaluation frameworks, and the crafting of data stories for the people who will exploit them. We also identify topics that are still worth investigating, such as the inclusion of different stakeholders’ profiles in designing data pipelines with the goal of providing data narration for all.","PeriodicalId":88813,"journal":{"name":"Advances in database technology : proceedings. International Conference on Extending Database Technology","volume":"56 1","pages":"855-858"},"PeriodicalIF":0.0,"publicationDate":"2023-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84774846","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}
This paper studies the publication of multi-dimensional data with local differential privacy (LDP). This problem raises tremendous challenges in terms of both computational efficiency and data utility. The state-of-the-art solution addresses this problem by first constructing a junction tree (a kind of probabilistic graphical model, PGM) to generate a set of noisy low-dimensional marginals of the input data and then using them to approximate the distribution of the input dataset for synthetic data generation. However, there are two severe limitations in the existing solution, i.e., calculating a large number of attribute pairs’ marginals to construct the PGM and not solving well in calculating the marginal distribution of large cliques in the PGM, which degrade the quality of synthetic data. To address the above deficiencies, based on the sparseness of the constructed PGM and the divisibility of LDP, we first propose an incremental learning-based PGM construction method. In this method, we gradually prune the edges (attribute pairs) with weak correlation and allocate more data and privacy budgets to the useful edges, thereby improving the model’s accuracy. In this method, we introduce a high-precision data accumulation technique and a low-error edge pruning technique. Second, based on joint distribution decomposition and redundancy elimination, we propose a novel marginal calculation method for the large cliques in the context of LDP. Extensive experiments on real datasets demonstrate that our solution offers desirable data utility.
{"title":"Multi-Dimensional Data Publishing With Local Differential Privacy","authors":"Gaoyuan Liu, Peng Tang, Chengyu Hu, Chongshi Jin, Shanqing Guo","doi":"10.48786/edbt.2023.15","DOIUrl":"https://doi.org/10.48786/edbt.2023.15","url":null,"abstract":"This paper studies the publication of multi-dimensional data with local differential privacy (LDP). This problem raises tremendous challenges in terms of both computational efficiency and data utility. The state-of-the-art solution addresses this problem by first constructing a junction tree (a kind of probabilistic graphical model, PGM) to generate a set of noisy low-dimensional marginals of the input data and then using them to approximate the distribution of the input dataset for synthetic data generation. However, there are two severe limitations in the existing solution, i.e., calculating a large number of attribute pairs’ marginals to construct the PGM and not solving well in calculating the marginal distribution of large cliques in the PGM, which degrade the quality of synthetic data. To address the above deficiencies, based on the sparseness of the constructed PGM and the divisibility of LDP, we first propose an incremental learning-based PGM construction method. In this method, we gradually prune the edges (attribute pairs) with weak correlation and allocate more data and privacy budgets to the useful edges, thereby improving the model’s accuracy. In this method, we introduce a high-precision data accumulation technique and a low-error edge pruning technique. Second, based on joint distribution decomposition and redundancy elimination, we propose a novel marginal calculation method for the large cliques in the context of LDP. Extensive experiments on real datasets demonstrate that our solution offers desirable data utility.","PeriodicalId":88813,"journal":{"name":"Advances in database technology : proceedings. International Conference on Extending Database Technology","volume":"18 1","pages":"183-194"},"PeriodicalIF":0.0,"publicationDate":"2023-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"86073013","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}
Panos Vassiliadis, Fation Shehaj, George Kalampokis, A. Zarras
In this paper, we address the problem of the co-evolution of Free Open Source Software projects with the relational schemata that they encompass. We exploit a data set of 195 publicly available schema histories of FOSS projects hosted in Github, for which we locally cloned their respective project and measured their evolution progress. Our first research question asks which percentage of the projects demonstrates a “hand-in-hand” schema and source code co-evolution? To address this question, we defined synchronicity by allowing a bounded amount of lag between the cumulative evolution of the schema and the entire project. A core finding is that there are all kinds of behaviors with respect to project and schema co-evolution, resulting in only a small number of projects where the evolution of schema and project progress in sync. Moreover, we discovered that after exceeding a 5-year threshold of project life, schemata gravitate to lower rates of evolution, which practically means that, with time, the schemata stop evolving as actively as they originally did. To answer a second question, on whether evolution comes early in the life of a schema, we measured how often does the cumulative progress of schema evolution exceed the respective progress of source change, as well as the respective progress of time. The results indicate that a large majority of schemata demonstrates early advance of schema change with respect to code evolution, and, an even larger majority is also demonstrating an advance of schema evolution with respect to time, too. Third, we asked at which time point in their lives do schemata attain a substantial
{"title":"Joint Source and Schema Evolution: Insights from a Study of 195 FOSS Projects","authors":"Panos Vassiliadis, Fation Shehaj, George Kalampokis, A. Zarras","doi":"10.48786/edbt.2023.03","DOIUrl":"https://doi.org/10.48786/edbt.2023.03","url":null,"abstract":"In this paper, we address the problem of the co-evolution of Free Open Source Software projects with the relational schemata that they encompass. We exploit a data set of 195 publicly available schema histories of FOSS projects hosted in Github, for which we locally cloned their respective project and measured their evolution progress. Our first research question asks which percentage of the projects demonstrates a “hand-in-hand” schema and source code co-evolution? To address this question, we defined synchronicity by allowing a bounded amount of lag between the cumulative evolution of the schema and the entire project. A core finding is that there are all kinds of behaviors with respect to project and schema co-evolution, resulting in only a small number of projects where the evolution of schema and project progress in sync. Moreover, we discovered that after exceeding a 5-year threshold of project life, schemata gravitate to lower rates of evolution, which practically means that, with time, the schemata stop evolving as actively as they originally did. To answer a second question, on whether evolution comes early in the life of a schema, we measured how often does the cumulative progress of schema evolution exceed the respective progress of source change, as well as the respective progress of time. The results indicate that a large majority of schemata demonstrates early advance of schema change with respect to code evolution, and, an even larger majority is also demonstrating an advance of schema evolution with respect to time, too. Third, we asked at which time point in their lives do schemata attain a substantial","PeriodicalId":88813,"journal":{"name":"Advances in database technology : proceedings. International Conference on Extending Database Technology","volume":"4 1","pages":"27-39"},"PeriodicalIF":0.0,"publicationDate":"2023-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90532072","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}
In constraint languages for RDF graphs, such as ShEx and SHACL, constraints on nodes and their properties are known as “shapes”. Using SHACL, we propose in this paper the notion of neighborhood of a node 𝑣 satisfying a given shape in a graph 𝐺 . This neighborhood is a subgraph of 𝐺 , and provides data provenance of 𝑣 for the given shape. We establish a correctness property for the obtained provenance mechanism, by proving that neighborhoods adhere to the Sufficiency requirement articulated for provenance semantics for database queries. As an additional benefit, neighborhoods allow a novel use of shapes: the extraction of a subgraph from an RDF graph, the so-called shape fragment. We compare shape fragments with SPARQL queries. We discuss implementation strategies for computing neighborhoods, and present initial experiments demonstrating that our ideas are fea-sible.
{"title":"Data Provenance for SHACL","authors":"Thomas Delva, Maxim Jakubowski","doi":"10.48786/edbt.2023.23","DOIUrl":"https://doi.org/10.48786/edbt.2023.23","url":null,"abstract":"In constraint languages for RDF graphs, such as ShEx and SHACL, constraints on nodes and their properties are known as “shapes”. Using SHACL, we propose in this paper the notion of neighborhood of a node 𝑣 satisfying a given shape in a graph 𝐺 . This neighborhood is a subgraph of 𝐺 , and provides data provenance of 𝑣 for the given shape. We establish a correctness property for the obtained provenance mechanism, by proving that neighborhoods adhere to the Sufficiency requirement articulated for provenance semantics for database queries. As an additional benefit, neighborhoods allow a novel use of shapes: the extraction of a subgraph from an RDF graph, the so-called shape fragment. We compare shape fragments with SPARQL queries. We discuss implementation strategies for computing neighborhoods, and present initial experiments demonstrating that our ideas are fea-sible.","PeriodicalId":88813,"journal":{"name":"Advances in database technology : proceedings. International Conference on Extending Database Technology","volume":"16 1","pages":"285-297"},"PeriodicalIF":0.0,"publicationDate":"2023-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90343818","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}
Chengcheng Wan, Yiwen Zhu, Joyce Cahoon, Wenjing Wang, K. Lin, Sean Liu, Raymond Truong, Neetu Singh, Alexandra Ciortea, Konstantinos Karanasos, Subru Krishnan
Database benchmarking and workload replay have been widely used to drive system design, evaluate workload performance, de-termine product evolution, and guide cloud migration. However, they both suffer from some key limitations: the former fails to capture the variety and complexity of production workloads; the latter requires access to user data, queries, and machine specifications, deeming it inapplicable in the face of user privacy concerns. Here we introduce our vision of learned workload synthesis to overcome these issues: given the performance profile of a customer workload (e.g., CPU/memory counters), synthesize a new workload that yields the same performance profile when executed on a range of hardware/software configurations. We present Stitcher as a first step towards realizing this vision, which synthesizes workloads by combining pieces from standard benchmarks. We believe that our vision will spark new research avenues in database workload replay.
{"title":"Stitcher: Learned Workload Synthesis from Historical Performance Footprints","authors":"Chengcheng Wan, Yiwen Zhu, Joyce Cahoon, Wenjing Wang, K. Lin, Sean Liu, Raymond Truong, Neetu Singh, Alexandra Ciortea, Konstantinos Karanasos, Subru Krishnan","doi":"10.48786/edbt.2023.33","DOIUrl":"https://doi.org/10.48786/edbt.2023.33","url":null,"abstract":"Database benchmarking and workload replay have been widely used to drive system design, evaluate workload performance, de-termine product evolution, and guide cloud migration. However, they both suffer from some key limitations: the former fails to capture the variety and complexity of production workloads; the latter requires access to user data, queries, and machine specifications, deeming it inapplicable in the face of user privacy concerns. Here we introduce our vision of learned workload synthesis to overcome these issues: given the performance profile of a customer workload (e.g., CPU/memory counters), synthesize a new workload that yields the same performance profile when executed on a range of hardware/software configurations. We present Stitcher as a first step towards realizing this vision, which synthesizes workloads by combining pieces from standard benchmarks. We believe that our vision will spark new research avenues in database workload replay.","PeriodicalId":88813,"journal":{"name":"Advances in database technology : proceedings. International Conference on Extending Database Technology","volume":"108 1","pages":"417-423"},"PeriodicalIF":0.0,"publicationDate":"2023-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"91107488","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}
X. Liu, Xu Cheng, Yanyan Yang, Huan Huo, Yongping Liu, P. S. Nielsen
Understanding crowd behavior is crucial for energy demand-side management. In this paper, we employ the fluid dynamics concept potential flow to model the energy demand shift patterns of the crowd in both temporal and spatial dimensions. To facilitate the use of the proposed method, we implement a visual analysis platform that allows users to interactively explore and interpret the shift patterns. The effectiveness of the proposed method will be evaluated through a hands-on experience with a real case study during the conference demonstration.
{"title":"Understanding crowd energy consumption behaviors","authors":"X. Liu, Xu Cheng, Yanyan Yang, Huan Huo, Yongping Liu, P. S. Nielsen","doi":"10.48786/edbt.2023.68","DOIUrl":"https://doi.org/10.48786/edbt.2023.68","url":null,"abstract":"Understanding crowd behavior is crucial for energy demand-side management. In this paper, we employ the fluid dynamics concept potential flow to model the energy demand shift patterns of the crowd in both temporal and spatial dimensions. To facilitate the use of the proposed method, we implement a visual analysis platform that allows users to interactively explore and interpret the shift patterns. The effectiveness of the proposed method will be evaluated through a hands-on experience with a real case study during the conference demonstration.","PeriodicalId":88813,"journal":{"name":"Advances in database technology : proceedings. International Conference on Extending Database Technology","volume":"6 1","pages":"799-802"},"PeriodicalIF":0.0,"publicationDate":"2023-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"91288051","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}
Ludovic Javet, N. Anciaux, Luc Bouganim, Léo Lamoureux, P. Pucheral
Can we push Edge computing one step further? This demonstration paper proposes an answer to this question by leveraging the generalization of Trusted Execution Environments at the very edge of the network to enable resilient and privacy-preserving computation on personal devices. Based on preliminary published results, we show that this can drastically change the way distributed processing over personal data is conceived and achieved. The platform presented here demonstrates the pertinence of the approach through execution scenarios integrating heterogeneous secure personal devices.
{"title":"Pushing Edge Computing one Step Further: Resilient and Privacy-Preserving Processing on Personal Devices","authors":"Ludovic Javet, N. Anciaux, Luc Bouganim, Léo Lamoureux, P. Pucheral","doi":"10.48786/edbt.2023.77","DOIUrl":"https://doi.org/10.48786/edbt.2023.77","url":null,"abstract":"Can we push Edge computing one step further? This demonstration paper proposes an answer to this question by leveraging the generalization of Trusted Execution Environments at the very edge of the network to enable resilient and privacy-preserving computation on personal devices. Based on preliminary published results, we show that this can drastically change the way distributed processing over personal data is conceived and achieved. The platform presented here demonstrates the pertinence of the approach through execution scenarios integrating heterogeneous secure personal devices.","PeriodicalId":88813,"journal":{"name":"Advances in database technology : proceedings. International Conference on Extending Database Technology","volume":"46 1","pages":"835-838"},"PeriodicalIF":0.0,"publicationDate":"2023-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90898008","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}
Loredana Caruccio, Stefano Cirillo, V. Deufemia, G. Polese, R. Stanzione
Query relaxation aims to relax the query constraints in order to derive some approximate results when the answer set is small. In this demo paper, we present REQUIRED, an automatized, portable, and scalable query relaxation tool leveraging metadata learned from an input dataset. The intuition is to use relationships underlying attribute values to derive a new query whose approximate results still meet the user’s expectations. In particular, REQUIRED exploits relaxed functional dependencies to modify the original query in two different ways: ( 𝑖 ) relaxing some query conditions by replacing the equality constraints with ranges and/or collections of admissible values, and ( 𝑖𝑖 ) rewriting the original query by replacing some or all the attributes involved in the conditions of the query with attributes related to them. Our demonstration scenarios show that REQUIRED is effective in properly relaxing queries according to the considered strategy.
{"title":"REQUIRED: A Tool to Relax Queries through Relaxed Functional Dependencies","authors":"Loredana Caruccio, Stefano Cirillo, V. Deufemia, G. Polese, R. Stanzione","doi":"10.48786/edbt.2023.74","DOIUrl":"https://doi.org/10.48786/edbt.2023.74","url":null,"abstract":"Query relaxation aims to relax the query constraints in order to derive some approximate results when the answer set is small. In this demo paper, we present REQUIRED, an automatized, portable, and scalable query relaxation tool leveraging metadata learned from an input dataset. The intuition is to use relationships underlying attribute values to derive a new query whose approximate results still meet the user’s expectations. In particular, REQUIRED exploits relaxed functional dependencies to modify the original query in two different ways: ( 𝑖 ) relaxing some query conditions by replacing the equality constraints with ranges and/or collections of admissible values, and ( 𝑖𝑖 ) rewriting the original query by replacing some or all the attributes involved in the conditions of the query with attributes related to them. Our demonstration scenarios show that REQUIRED is effective in properly relaxing queries according to the considered strategy.","PeriodicalId":88813,"journal":{"name":"Advances in database technology : proceedings. International Conference on Extending Database Technology","volume":"10 1","pages":"823-826"},"PeriodicalIF":0.0,"publicationDate":"2023-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"86751714","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}
Nils Strassenburg, Dominic Kupfer, J. Kowal, T. Rabl
Deep learning models are deployed in an increasing number of industrial domains, such as retail and automotive applications. An instance of a model typically performs one specific task, which is why larger software systems use multiple models in parallel. Given that all models in production software have to be managed, this leads to the problem of managing sets of related models, i.e., multi-model management. Existing approaches perform poorly on this task because they are optimized for saving single large models but not for simultaneously saving a set of related models. In this paper, we explore the space of multi-model management by presenting three optimized approaches: (1) A baseline approach that saves full model representations and minimizes the amount of saved metadata. (2) An update approach that reduces the storage consumption compared to the baseline by saving parameter updates instead of full models. (3) A provenance approach that saves model provenance data instead of model parameters. We evaluate the approaches for the multi-model management use cases of managing car battery cell models and image classification models. Our results show that the baseline outperforms existing approaches for save and recover times by more than an order of magnitude and that more sophisticated approaches reduce the storage consumption by up to 99%.
{"title":"Efficient Multi-Model Management","authors":"Nils Strassenburg, Dominic Kupfer, J. Kowal, T. Rabl","doi":"10.48786/edbt.2023.37","DOIUrl":"https://doi.org/10.48786/edbt.2023.37","url":null,"abstract":"Deep learning models are deployed in an increasing number of industrial domains, such as retail and automotive applications. An instance of a model typically performs one specific task, which is why larger software systems use multiple models in parallel. Given that all models in production software have to be managed, this leads to the problem of managing sets of related models, i.e., multi-model management. Existing approaches perform poorly on this task because they are optimized for saving single large models but not for simultaneously saving a set of related models. In this paper, we explore the space of multi-model management by presenting three optimized approaches: (1) A baseline approach that saves full model representations and minimizes the amount of saved metadata. (2) An update approach that reduces the storage consumption compared to the baseline by saving parameter updates instead of full models. (3) A provenance approach that saves model provenance data instead of model parameters. We evaluate the approaches for the multi-model management use cases of managing car battery cell models and image classification models. Our results show that the baseline outperforms existing approaches for save and recover times by more than an order of magnitude and that more sophisticated approaches reduce the storage consumption by up to 99%.","PeriodicalId":88813,"journal":{"name":"Advances in database technology : proceedings. International Conference on Extending Database Technology","volume":"77 1","pages":"457-463"},"PeriodicalIF":0.0,"publicationDate":"2023-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"86764458","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：