Amazing technological breakthrough possible @S-Logix pro@slogix.in

Office Address

  • #5, First Floor, 4th Street Dr. Subbarayan Nagar Kodambakkam, Chennai-600 024 Landmark : Samiyar Madam
  • pro@slogix.in
  • +91- 81240 01111

Social List

Incremental Learning for Large-Scale Data Stream Analytics in a Complex Environment

Incremental Learning for Large-Scale Data Stream Analytics in a Complex Environment

Top PhD Thesis on Incremental Learning for Large-Scale Data Stream Analytics in a Complex Environment

Research Area:  Machine Learning

Abstract:

Data streams are not only characterized by continuous and non-stationary characteristics, butthey also arrive at a rapid rate from multiple sources, generating high volumes of data whichmight be beyond the capacity of single processing node machine learning. While most evolvingalgorithms are adaptive, have an open structure and operate in a single-pass learning mode, theyare not designed to run in a distributed processing environment. A novel distributed evolvingsystem, a large-scale data stream analytics framework based on a parsimonious network of fuzzyinference system (PANFIS), termed Scalable PANFIS, is proposed. Scalable PANFIS utilizesthe PANFIS evolving algorithm distributed across Spark computing nodes to train large-scaledata streams. Scalable PANFIS can deal with large-scale examples, generating a fast-evolvingdistributed model using an expertly designed merging model mechanism.Furthermore, the active learning (AL) method is designed to run with PANFIS to accelerate thePANFIS learning mechanism further. We design four-structure Scalable PANFIS algorithmsusing a combination of PANFIS with/without AL and the merging/majority voting method.The results show that all these combinations produce comparable accuracy across all datasets.Of these four structures, the combination of merging models with AL trains large-scale datasetswith a significantly faster running time. In comparison to the other algorithms in the Sparklibrary, Scalable PANFIS methods yield higher accuracy, despite a slightly slower training timefor some cases.

Name of the Researcher:  Za in, Choiru

Name of the Supervisor(s):  Dr. Mahardhika Pratama, Dr. Eric Pardede, andDr. Zhen He

Year of Completion:  2021

University:  La Trobe University

Thesis Link:   Home Page Url