Table 7

Algorithm: Mean field annealing

\SetKwInOut{Input}{Input}

\SetKwInOut{Output}{Output}

\Input{A set of observations (t _ij , s _ij ) for each pair (i, j), ψ, a number of clusters K}

\Output{A probability q _ik for a node i belonging to a cluster k for all i and for all k}

Initialize q to random values;

Initialize annealing factor γ;

While γ <γ_max Repeat q converges

ForAll i ∈ V

ForAll k ∈ K

$C_{ik}={\displaystyle \sum _{j\in Neighbor(i)}{q}_{jk}(t_{ij}-s_{ij})+(1-q_{jk})\psi s_{ij}}$

ForAll k ∈ K

${\widehat{q}}_{ik}=\frac{\exp (-\gamma C_{ik})}{{\displaystyle \sum _{l=1}^K\exp (-\gamma C_{il})}}$

ForAll k ∈ K

q _ik = ${\widehat{q}}_{ik}$

Increase γ;