MindSpore Probability Documents

A deep learning model has a strong fitting capability, and the Bayesian theory has a good explainability. MindSpore probabilistic programming provides a framework that seamlessly integrates Bayesian learning and deep learning. It aims to provide users with a complete probability learning library for establishing probabilistic models and applying Bayesian inference.

Probabilistic programming provides the following functions:

  • Abundant statistical distribution and common probabilistic inference algorithms

  • Combinable probabilistic programming modules for developers to use the logic of the deep learning model to build a deep probabilistic model

  • A toolbox for uncertainty estimation and anomaly detection to extend Bayesian applications

Typical MindSpore Probability Application Scenarios

  1. Building the Bayesian Neural Network

    Use the Bayesian neural network to classify images.

  2. Building the Variational Autoencoder

    Use the variational autoencoder to compress the input data to generate new samples.

  3. One-click Conversion from DNN to BNN

    Convert DNN models into BNN models with one click.

  4. Using the Uncertainty Evaluation Toolbox

    Use the uncertainty evaluation toolbox to obtain the accidental uncertainty and cognitive uncertainty and to better understand models and datasets.

Guide

API References