Have any of you ever found interesting use cases where Bayesian methods (MCMC, HMM, etc.) outperform other methods (ARIMA, RNN, XGB, logistic regression, etc.) for forecasting purposes (time series or otherwise)? Really curious about stochastic processes in general, but not sure it's worth the time investment to get up to speed at this point. Appreciate any stories you can share!
What’s the diff between inference and prodiction?
inference = understanding the underlying statistical behavior of a dataset. prediction = estimating new data
its not uncommon to use mcmc techniques in astrophysics. look at the projects that use the emcee library. i’ve used it doing reconstructions of historical data in the weather space
Yes. My wife worked in exactly this subfield during her Math PhD, and her work was fascinating and exciting.
Hmms map are some of the basic algorithms at play in most of your mobile devices. Google maps, wireless modems use bayesian estimation in receivers design. This whole ml ai fad failed to highlight that these algos have been in real world for a while more than 40+ years.
for a jumping off point check out the viterbi algorithm. the creator went on to found qualcomm amongst other things
They found the lost Air France flight using Bayesian methods
I use hierarchical Bayesian methods plus decision theory all the time in the medical informatics field. High noise problems are the sweet spot if you need to characterize uncertainties. Also Bayesian kalman filters for time series etc...
Market mix modelling
bayesian methods usually used for inference not predictions right? i've heard they can be good with small data where distributions are known