Restricted Boltzmann Machines are Hard to Approximately Evaluate or Simulate

87 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

A practical tutorial on autoencoders for nonlinear feature fusion: Taxonomy, models, software and guidelines (2018, background): “However, computing probabilities is unfeasible since the distribution is intractable, and they have been proved to be hard to simulate [109].”
Discrete Variational Autoencoders (2016, background): “2 However, sampling from undirected graphical models is generally intractable [14], as is sampling from the posterior of a directed graphical model conditioned on its leaf variables [15].”
Quantum machine learning: a classical perspective (2017, background): “It has been shown that computing the log-likelihood and sampling from an RBM is computationally hard [131].”
Universal Approximation Depth and Errors of Narrow Belief Networks with Discrete Units (2013, background): “periments evaluating the divergence from probability models deﬁned by neural networks are only feasible for small networks, since otherwise the model distributions are too hard to compute (see, e.g., Long and Servedio, 2010). For large models one still could try to sample the distributions and repla”
Relaxations for inference in restricted Boltzmann machines (2013, background): “The latter is discussed and motivated by Salakhutdinov & Murray (2008), and is generally interesting as it captures the essential theoretical hardness of RBM inference (Long & Servedio, 2010).”