p(w|α) = N(w|0, α−1I), maximization of the corresponding posterior p(w|t) with respect to w is equivalent to the minimization of β 2 N ∑ n=1 { tn − w⊤ϕ(xn) } 2 + α 2 w⊤w (3.55’) the minimization corresponds to (3.27) with λ = α/β. 7/9
p(w|α) = N(w|0, α−1I), maximization of the corresponding posterior p(w|t) with respect to w is equivalent to the minimization of β 2 N ∑ n=1 { tn − w⊤ϕ(xn) } 2 an error function + α 2 w⊤w (3.55’) the minimization corresponds to (3.27) with λ = α/β. 7/9
p(w|α) = N(w|0, α−1I), maximization of the corresponding posterior p(w|t) with respect to w is equivalent to the minimization of β 2 N ∑ n=1 { tn − w⊤ϕ(xn) } 2 an error function + α 2 w⊤w a quadratic regularization (3.55’) the minimization corresponds to (3.27) with λ = α/β. 7/9
as a known constant. Where the likelihood function of t is defined as: p(t|w) = N ∏ n=1 N(tn|w⊤ϕ(xn), β−1) The exponential of a quadratic func. of w (3.10’) 8/9