Design of Stochastic Machines Dedicated to Approximate Bayesian inferences

Marvin Faix (1, 2, 3) , Raphä El Laurent (4) , Pierre Bessière (5, 6, 7) , Emmanuel Mazer (1, 5) , Jacques Droulez (6, 5, 7)

1 PERVASIVE INTERACTION - PERVASIVE INTERACTION
2 UGA [2016-2019] - Université Grenoble Alpes [2016-2019]
3 LIG - Laboratoire d'Informatique de Grenoble
4 Probayes [Montbonnot]
5 CNRS - Centre National de la Recherche Scientifique
6 ISIR - Institut des Systèmes Intelligents et de Robotique
7 AMAC

We present an architecture and a compilation toolchain for stochastic machines dedicated to Bayesian inferences. These machines are not Von Neumann and code information with stochastic bitstreams instead of using floating point representations. They only rely on stochastic arithmetic and on Gibbs sampling to perform approximate inferences. They use banks of binary random generators which capture the prior knowledge on which the inference is built. The output of the machine is devised to continuously sample the joint probability distribution of interest. While the method is explained on a simple example, we show that our machine computes a good approximation of the solution to a problem intractable in exact inference.

Mots clés: stochastic computing; Bayesian Programming

Domaines: Informatique; Intelligence artificielle [cs.AI]; Architectures Matérielles [cs.AR]

Marvin Faix, Raphä El Laurent, Pierre Bessière, Emmanuel Mazer, Jacques Droulez. Design of Stochastic Machines Dedicated to Approximate Bayesian inferences. IEEE Transactions on Emerging Topics in Computing, 2016, ⟨10.1109/TETC.2016.2609926⟩. ⟨hal-01374906⟩