… log-supermodular distributions, and we also cover challenges and suitability of parameter
learning … We will outline log-supermodular models which are crucial for this thesis, and discuss …

… estimation for estimating parameters for continuous log-supermodular distributions, which is
… We then show how “perturb-and-MAP” ideas based on the Gumbel distribution and efficient …

… parameter that must be learned. We leave the question of how best to learn these parameters
… Scalable variational inference in log-supermodular models. In International Conference on …

… where the time steps are partitioned (unbeknownst to the algorithm) into an adversarial part
and a stochastic part, where in the latter jobs are generated iid from an unknown distribution. …

… distributions that imply strong forms of positive dependence; namely log-supermodular (MTP
2 ) distributions … they do not impose any parametric assumptions. Popular such methods …

… Both learning the model parameters θ from data, as well as quantifying uncertainty and …
resulting distributions of this form are also referred to as log-submodular and log-supermodular …

… parameters, w ≥ 0, are given as the output of the parameter … a log-concave or log-convex
distribution, but in general, it will … We note that we will restrict log-supermodular models in the …

… parameter in P ( x ) (Eq. (7)) by using E ( x ) ∕ T and decrease T , the resulting distribution
… We point out that the notion of a temperature parameter T in a probability distribution and …

… parameter f(.) that measures the badness of the structure (or if there is a given parameter f(.)
already, then we prepare a new parameter f… If the expected value of this badness parameter …

… objective over a set of parameters where the objective is also … of (|) which allows the use of
the distributed law, and this is one of the … Log-supermodular models give high probability to …