Probit Approximation

  • avoid long computation time
  • assume likelihood is also normally distributed

p(w|D)=N(w|μ,Σ)

Approximation of Posterior

  • sigmoid is similar to normal CDF

σ(a)Φ(aπ8)

  • approx posterior via sigmoid

p(y=1|x,D)=σ(m1+πv8)m=E[a]=xTμv=V[a]=xTΣxa=xTw

  • produces estimates that are closer to the decision boundary