• Used to calculate distances or similarity between vectors
  • p-norm
  • For vectors

Comparisions

  • Robustness
    • resistance to outliers
    • L1 > L2
      • the L2 norm squares values, so it increases the cost of outliers exponentially
  • Stability
    • resistance to horizontal adjustments.
    • L2 > L1
  • Computational Difficulty
    • L2 > L1
    • L2 has a closed form solution
    •  L1 does not have a closed form solution because it is a non-differenciable piecewise function, as it involves an absolute value. For this reason, L1 is computationally more expensive, as we can’t solve it in terms of matrix math
  • Sparsity
    • L1 > L2

Matrices