Efficient Construction of Nonlinear Models over Normalized Data

Cheng, Zhaoyue; Koudas, Nick; Zhang, Zhe; Yu, Xiaohui

doi:10.1109/icde51399.2021.00103

Cited by 3 publications

(3 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recent advances [37], [38], [69] facilitate more automation and optimization opportunities by decomposing ML models to basic building blocks of linear algebra (LA) operators, i.e., developing rewriting rules for LA operators and then pushing them down to joinable tables. Existing solutions for factorization over joins [37], [70], [71] mainly tackle inner joins. Our main contribution is to expand the dataset relationships to more integration scenarios, with left joins, full outer joins, and unions, cf.…”

Section: Algebraic Computation Over Silosmentioning

confidence: 99%

“…Another interesting direction is to support more ML models, e.g., transformers [72]. In existing works, non-linearity is studied with regard to feature interactions [70] and ML models, e.g., Gaussian Mixture Models, Neural Networks [71].…”

Section: A Computation Challenge: DI Metadata For Factorizationmentioning

confidence: 99%

“…In general terms, if by joining the source tables we obtain a target table with more instance redundancy than source tables, factorization may be faster than materialization. To distinguish the Area I in Figure 6 from others, two heuristic metrics are proposed: tuple ratio (TR) and column (or feature) ratio (FR) [37], [70], [74]. The TR, representing duplicated rows postjoin, is defined as #tuples of S 2 #tuples of S 1 .…”

Section: B Cost Estimation Challenge: To Factorize or To Materializementioning

confidence: 99%

See 2 more Smart Citations

Amalur: The Convergence of Data Integration and Machine Learning

Li,

Sun,

Zhan

et al. 2024

IEEE Trans. Knowl. Data Eng.

View full text Add to dashboard Cite

Machine learning (ML) training data is often scattered across disparate collections of datasets, called data silos. This fragmentation poses a major challenge for data-intensive ML applications: integrating and transforming data residing in different sources demand a lot of manual work and computational resources. With data privacy constraints, data often cannot leave the premises of data silos; hence model training should proceed in a decentralized manner. In this work, we present a vision of how to bridge the traditional data integration (DI) techniques with the requirements of modern machine learning systems. We explore the possibilities of utilizing metadata obtained from data integration processes for improving the effectiveness, efficiency, and privacy of ML models. Towards this direction, we analyze ML training and inference over data silos. Bringing data integration and machine learning together, we highlight new research opportunities from the aspects of systems, representations, factorized learning, and federated learning.

show abstract

Section: Algebraic Computation Over Silosmentioning

confidence: 99%

Section: A Computation Challenge: DI Metadata For Factorizationmentioning

confidence: 99%

Section: B Cost Estimation Challenge: To Factorize or To Materializementioning

confidence: 99%

See 1 more Smart Citation

Amalur: The Convergence of Data Integration and Machine Learning

Li,

Sun,

Zhan

et al. 2024

IEEE Trans. Knowl. Data Eng.

View full text Add to dashboard Cite

show abstract

Amalur: Data Integration Meets Machine Learning

Hai,

Koutras,

Ionescu

et al. 2023

2023 IEEE 39th International Conference on Data Engineering (ICDE)

View full text Add to dashboard Cite

Improved Approximation Algorithms for Relational Clustering

Esmailpour,

Sintos

2024

Proc. ACM Manag. Data

View full text Add to dashboard Cite

Clustering plays a crucial role in computer science, facilitating data analysis and problem-solving across numerous fields. By partitioning large datasets into meaningful groups, clustering reveals hidden structures and relationships within the data, aiding tasks such as unsupervised learning, classification, anomaly detection, and recommendation systems. Particularly in relational databases, where data is distributed across multiple tables, efficient clustering is essential yet challenging due to the computational complexity of joining tables. This paper addresses this challenge by introducing efficient algorithms for k-median and k-means clustering on relational data without the need for pre-computing the join query results. For the relational k-median clustering, we propose the first efficient relative approximation algorithm. For the relational k-means clustering, our algorithm significantly improves both the approximation factor and the running time of the known relational k-means clustering algorithms, which suffer either from large constant approximation factors, or expensive running time. Given a join query q and a database instance D of O(N) tuples, for both k-median and k-means clustering on the results of q on D, we propose randomized (1+ε)γ-approximation algorithms that run in roughly O(k 2 N fhw )+T_γ(k 2 ) time, where ε ∈ (0,1) is a constant parameter decided by the user, \fhw is the fractional hyper-tree width of Q, while γ and T_γ(x) represent the approximation factor and the running time, respectively, of a traditional clustering algorithm in the standard computational setting over x points.

show abstract

Efficient Construction of Nonlinear Models over Normalized Data

Cited by 3 publications

References 23 publications

Amalur: The Convergence of Data Integration and Machine Learning

Amalur: The Convergence of Data Integration and Machine Learning

Amalur: Data Integration Meets Machine Learning

Improved Approximation Algorithms for Relational Clustering

Contact Info

Product

Resources

About