search

piEsposito / blitz-bayesian-deep-learning

A simple and extensible library to create Bayesian Neural Network layers on PyTorch.
GNU General Public License v3.0
929 stars 105 forks source link

Uncertainties on input data #73

Closed JSM-dfm closed 3 years ago

JSM-dfm commented 3 years ago

Hello,

first off thank you for sharing this project!

I am looking for a way to include uncertainties in my input data. Basically, I want to do a regression on experimental data, where some of the data is more accurately determined, and therefore I want it to have a larger "weight".

I looked through the examples and could not find a way to do this.

First, I want to confirm if what I want is possible. Secondly, it would be most appreciated if anyone could point me in the right direction.

Thank you for your time reading this :)

piEsposito commented 3 years ago

You can just add noise or do some data agumentation on your input. Regarding the weights, you can do that by weighting more your "good" data on the loss, tho I don't think that approach will be very useful.