Bayesian Methods offer a statistically-robust route to assessing how confident one should be in the predictions of a machine learning model. Knowing how to work with Bayesian methods is a excellent when you have small amounts of data, a good sense of the model which could describe your data, or need for meaningful error bars on the properties.
The learning objectives for our model are:
- Learning how to define the prior and likelihood functions. How do you include the model form in the likehood function?
- Understanding MCMC sampling. What is burn-in and how do I achieve it?
- Comparing the evidence for different models. What is a Bayes factor and what is a large Bayes factor?
- Using the posterior to estimate error bars on predictions. Can you produce a 95% confidence interval?
A few papers to read to gain a better understanding of this field include:
- Paulson et al. IJES (2019): Tutorial and demonstration of using Bayesian methods to predict thermoelectric properties of materials
- de Schoot et al. Nature Reviews (2021): Great tutorial review of methods behind Bayesian models