Bayesian Neural Network

• Bayesian Model Estimation
• Generally we want to learn Joint Probability distribution $P(y|x)$ but this does not use the model parameters w
• We need $P(w|D)=\frac{P(D|w)P(w)}{P(D)}$
  • D is the labelled dataset
  • Model is now defined by structure and parameters
• The parameters encode information about Uncertainty
  • Can be understood using Bayesian Predictive Posterior