Hacker News new | past | comments | ask | show | jobs | submit login

> Is Bayesian Deep Learning worth looking into if I don't have much interest in statistical applications but much more just using deep learning as general purpose functions?

If you manage to build a model with a high accuracy and a sensible uncertainty on the output, you can use that information to do a lot of great things such as :

  - obviously apply the method to domains that require uncertainty for legal or technical reasons (simulation)
  - adding sample to improve you knowledge around uncertain input (active learning)
  - use an optimized betting strategy that takes risks into account (bayesian optimization)
Gaussian process are a prime example of that but I am not aware of a deep learning approach that realises this (yet).



I stand corrected, after some research, I found a paper that offers a promising way to get good uncertainty information from a deep learning model : http://www.cs.ox.ac.uk/people/yarin.gal/website/blog_3d801aa...




Applications are open for YC Summer 2020

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: