Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Fuzzy collaborative forecasting and clustering : methodology, system architecture, and applications
Chen T., Honda K., Springer International Publishing, New York, NY, 2019. 100 pp. Type: Book (978-3-030225-73-5)
Date Reviewed: Jun 14 2021

Collaborative machine learning (ML)--sometimes known as federated ML--has been gaining momentum in the last few years for several in-demand reasons, including privacy-preserving data analysis, sharing complex computations (such as those required for model training) in low-resource devices like smartphones, and so on. This slim book by Chen and Honda provides readers with new developments in collaborative forecasting and clustering through techniques based on fuzzy logic.

Collaborative ML enables the generation of data-driven models by involving several computing systems (called “sites” in the book) that may not share data. When data sharing is not allowed, and data represent rows of objects described by columns of features, two possible scenarios are considered: (i) each site holds a subset of objects represented by all features (what is called “horizontally distributed databases”); or (ii) each site holds a database of all objects, but described by a subset of features only (in the case of vertically distributed databases). More complex situations arise when features belong to the same domain--in such a case, they are called “items”--thus enabling co-clustering in the object and item domains simultaneously to find co-occurrence relations. (Co-clustering emerges in fields such as bioinformatics, information retrieval, and so on, and thus it is a relevant topic of current research.) Even more complex problems are described by three-mode datasets where, according to the book, two datasets are defined: a dataset relating objects and items, and a second dataset relating items and “ingredients.” In this case, collaboration can take place when the two datasets are available in two separate sites and are not shared between the two.

The authors tackle all of these situations and summarize the main results of their research in this book. After some preliminary notions, the book opens with some methods for collaborative forecasting using fuzzy linear regression (FLR), fuzzy analytic hierarchy processes (FAHP), and fuzzy back-propagation networks (BPN). In all these cases, the objective is to reach a consensus among agents who are using either the same schema (for example, all agents use FLR) but with different parameterization, or different schemes (for example, some agents use FLR, others FBN, and so on) provided that outputs are compatible (such as triangular fuzzy numbers). A number of solutions are proposed to guide agents to reach a consensus (that is, to modify the parameters of their models so that their outputs come closer to the other agents according to a defined objective function), as well as techniques for merging the final outputs of all agents and the consequent defuzzification (that is, the synthesis of a fuzzy set into a scalar number) of the result, if required.

The second part of the book is devoted to collaborative fuzzy clustering and fuzzy co-clustering (with both classical and three-mode datasets). In both horizontally and vertically distributed datasets, the proposed collaboration techniques are based on disentangling the contribution of each site in the objective function of the clustering problem. Some techniques are defined so that sites can communicate aggregate information only (for example, cluster prototypes) that is eventually used by other sites for their clustering tasks, which are usually modified in order to manage both local data and aggregate information coming from the other sites. In some situations, information obfuscation is required to avoid uncovering private data; in such cases, a simple yet effective solution (devised by Vaidya and Clifton [1]) is applied because of the numerical nature of handled data.

The book and included research results are highly specialized. However, as it is organized in a coherent way, it is inspiring for further research on the topic. Due to the slim structure of the book, the state of the art is not covered save some bibliographical references. Also, some parts are discussed a bit too hastily, such as in the case of determining the authority of agents, which would be of great interest beyond the specific collaboration techniques. Finally, some real-world examples are missing, which would show the real effectiveness of the described methods. Overall, the book is recommended for researchers in collaborative ML who want a quick look at collaboration mechanisms via special prediction and clustering techniques based on fuzzy logic.

Reviewer:  Corrado Mencar Review #: CR147285 (2109-0231)
1) Vaidya, J. Clifton, C. Privacy-preserving k-means clustering over vertically partitioned data. In Proc. of the Ninth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining ACM, 2003, 206–215.
Bookmark and Share
  Featured Reviewer  
 
Fuzzy Set (I.5.1 ... )
 
 
Clustering (H.3.3 ... )
 
 
Collaborative Computing (H.5.3 ... )
 
 
Vague (H.2.4 ... )
 
Would you recommend this review?
yes
no
Other reviews under "Fuzzy Set": Date
A supervised learning algorithm for hierarchical classification of fuzzy patterns
Biswas P., Majumdar A. Information Sciences 31(2): 91-106, 1983. Type: Article
Feb 1 1985
Estimation of fuzzy memberships from histograms
Devi B., Sarma V. Information Sciences 35(1): 43-59, 1985. Type: Article
Nov 1 1985
Fuzzy mathematical approach to pattern recognition
Pal S. (ed), Dutta-Majumder D., Halsted Press, New York, NY, 1986. Type: Book (9789780470274637)
Jun 1 1987
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy