In real-world applications, one has to predict unknown values (disease risks, stocks or assets values, etc.) based on given information. Along with state-of-art methods, we develop novel algorithms that allow one to not only predict the desired values, but also estimate the certainty with which the predictions hold. All the methods are validated on real-world data sets.

Principle Investigators: Pavel Gurevich, Hannes Stuke

**Publications**

Gurevich P., Stuke H.
**Gradient conjugate priors and deep neural networks.** Preprint: arXiv:1802.02643 [math.ST] (pdf) |

Gurevich P., Stuke H.** **
**Learning uncertainty in regression tasks by artificial neural networks. ** Preprint: arXiv:1707.07287 [stat.ML] (pdf) |