Week 2- Hybrid computing using a neural network with dynamic external memory (2)

original file link

  • Meeting Date: 11.07.2018
  • Attendee: Won, Mike, Chanoh

Paper reviewed

  • Paper Title: Hybrid computing using a neural network with dynamic external memory

Meeting summary

  • Reviewed all operations and their meaning.
  • PPT: here
  • Note: here
  • Operation sequence from a code ananysis
    • (Code link)[https://github.com/ixaxaar/pytorch-dnc]. In pytorch, easy to understand, works well.
    • Write
      • Find similar contents “c”
      • Address generation w = gw(ga*a+(1-ga)c)
        • See which memory cell is free
        • Delete or not “”
        • Important “” (decending order)
        • Usage vector “pi”
        • Allocated adress “a”
      • Reset
      • Memory writing “M”
      • Link update “L”
      • Precedence update “p”
    • Read
      • Find similar contents “c”
      • Address generation
        • Similar contents
        • Forward
        • Backward
      • Read
  • Helpful block diagram
  • Questions unanswered
    • What is oneplus??

Next Meeting

  • Meeting Date: 23.07.2018
  • Paper: Neural SLAM, Neural Map
    • https://arxiv.org/abs/1706.09520
    • https://arxiv.org/abs/1702.08360
  • Useful Video: https://vimeo.com/252185932
Written on July 11, 2018