Notes from feedback

• We can dispense with the warm-up (few liked it)
• Randomly shuffled groups are tricky
  • Will try nailing down groups
  • Survey: who do you want to work with?
• Rearranging group work breakout
  • Will try to post one (longer) notebook by start of week
  • Exercises associated with Tuesday + previous Thursday
  • Budget half an hour group at end of Thursday lecture
  • May end Tuesday lectures walking through notebook setup

Reminders:

• You will get more out of class if you read ahead!
• Ask questions early!

Constraints and multipliers

This will be terse, but I have some longer notes (notebook, pdf).

• KKT conditions and constrained optimization
• Physical interpretation of Lagrange multipliers
• Barriers, penalties, and projected gradients

NMF

• Basics
  • Definition, non-negative rank
  • Parts decomposition and sparsity
  • Applications
• Separable NMF and interpolative decomposition
  • Basic assumption
  • Pure pixels, anchor words, etc
  • Role of pivoted QR
• Frobenius norm NMF (vs other loss functions)
• Gradient-based algorithms
  • Projected gradient descent
  • Lee-Seung update
• (Block) coordinate descent methods
  • Simple coordinate descent
  • Block descent and HALS-RRI
  • ANLS and the non-negative LS building block