Uploaded image for project: 'Data Management'
  1. Data Management
  2. DM-17749

Investigate the use of Adam with proximal operators

    Details

    • Type: Story
    • Status: Invalid
    • Resolution: Done
    • Fix Version/s: None
    • Component/s: meas_deblender
    • Labels:
      None
    • Templates:
    • Story Points:
      6
    • Sprint:
      DRP F19-5
    • Team:
      Data Release Production

      Description

      It should be fairly straightforward to implement autograd into scarlet (DM-17748), however automatic gradients alone can only accomplish so much. The problem of calculating step sizes still remains and many accelerated methods that automatically calculate step sizes do not seem to work out of the box with proximal operators. This ticket is to investigate the Adam algorithm (Kingma and Ba 2015) to see if there is a way to make it robust in cases where we take proximal gradients, such as our optimization in scarlet.

        Attachments

          Activity

            People

            • Assignee:
              fred3m Fred Moolekamp
              Reporter:
              fred3m Fred Moolekamp
              Watchers:
              Fred Moolekamp
            • Votes:
              0 Vote for this issue
              Watchers:
              1 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved:

                Summary Panel