Pattern Recognition, 4th Edition

The AdaBoost Algorithm

• Initialize:
• Initialize: m = 1
• Repeat
  • Compute optimum θ^m in ϕ(·; θ^m) by minimizing P^m; (4.135)
  • Compute the optimum P^m; (4.135)
  • 
  • Z^m = 0.0
  • For i = 1 to N
    • w^{(m + 1)i} = w^(m)i exp(-yⁱα^mϕ(xⁱ; θ^m))
    • Z^m = Z^m + w^{(m + 1)i}
  • End{For}
  • For i = 1 to N
    • w^{(m + 1)i} = w^{(m + 1)i}/Z^m
  • End {For}
  • K = m
  • m = m + 1
• Until a termination criterion is met.
• 

One of the main and very interesting properties of boosting is its relative immunity to ...