VINE: A Variational Inference -Based Bayesian Neural Network Engine
Technical Report,01 Oct 2016,01 Aug 2017
University of Southern California Los Angeles United States
Pagination or Media Count:
This report describes our findings and results for the DARPA MTO seedling project titled SpiNN-SC Stochastic Computing-Based Realization of Spiking Neural Networks also known as VINE A Variational Inference-Based Bayesian Neural Network Engine. The primary goal was to develop a Bayesian Neural Network BNN with an integrated Variational Inference VI engine to perform inference and learning statically and on-the-fly under uncertain or incomplete input and output features. A secondary goal is to enable robust decision making under noise and variability in the observed data and without reference to a ground truth. The key expected impact is to enable a new generation of BNNs that can operate on input and output features specified as random variables, that admit efficient hardware realization, and that can not only do inference but also can be retrained on-the-fly based on incoming data.