Learning in Compressed Domain Development of Light Learning Engine in Fog Computing Environments

Main Achievement:


–     Low-complexity compressed analysis (CA)

  • Analyze signal in compressed domain with no reconstruction

–     Framework:

1. signal processing directly on compressed signal

–     Detection with two-stage class-dependent dictionary

–     Compressive analysis with PCA-assisted dictionary

–     Robust CA using Subspace-based Dictionary

»     Compressive interference removal

2. light machine learning model for distributed system

–     Aggregation of light classifier ELM




Quantitative Impact: 


–      Significantly reduce complexity by removing reconstruction

  • 94.89% reduced in training
  • 99.77% reduced in analysis

End Goal:

–     Compressive E-Health Tracking for Quantified Well-Being

–     Compressed analysis using Subspace-based dictionary with light distributed model 

–     Implement proposed CA engine on distributed/heterogeneous platform with hardware acceleration

 (Updated in Jul, 2017)