Projects funded by the NCN


Information on the principal investigator and host institution

Information of the project and the call

Keywords

Equipment

Delete all

Balancing priors and learning biases to improve Bayesian Neural Networks

2022/45/P/ST6/02969

Keywords:

Bayesian neural networks approximate inference prior learning uncertainty quantification

Descriptors:

  • ST6_007:
  • ST6_011:

Panel:

ST6 - Computer science and informatics: informatics and information systems, computer science, scientific computing, intelligent systems

Host institution :

Uniwersytet Jagielloński, Wydział Matematyki i Informatyki

woj. małopolskie

Other projects carried out by the institution 

Principal investigator (from the host institution):

dr Tomasz Kuśmierczyk 

Number of co-investigators in the project: 5

Call: POLONEZ BIS 2 - announced on 2022-03-28

Amount awarded: 1 078 901 PLN

Project start date (Y-m-d): 2024-04-01

Project end date (Y-m-d): 2026-03-31

Project duration:: 24 months (the same as in the proposal)

Project status: Pending project

Project description

Download the project description in a pdf file

Note - project descriptions were prepared by the authors of the applications themselves and placed in the system in an unchanged form.

Information in the final report

  • Articles in post-conference publications (3)
  1. Hypernetwork Approach to Bayesian MAML
    Authors:
    Piotr Borycki, Piotr Kubacki, Marcin Przewięźlikowski, Tomasz Kuśmierczyk, Jacek Tabor, Przemysław Spurek
    Conference:
    Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 39 No. 28: IAAI-25, EAAI-25, AAAI-25 Student Abstracts, Undergraduate Consortium and Demonstrations (rok: 2025, tom: The Thirty-Ninth AAAI Conference on Artificial Intelligence (AAAI-25), strony: 29325-29327), Wydawca: AAAI Press
    Data:
    konferencja February 27 – March 4, 2025
    Status:
    Published
    DOI:
    10.1609/aaai.v39i28.35239 - link to the publication
  2. Hi-fi functional priors by learning activations
    Authors:
    Marcin Sendera, Amin Sorkhei, Tomasz Kuśmierczyk
    Conference:
    Workshop publications are not included in the offical proceedings of the NeurIPS conference, but are published online through OpenReview (rok: 2024, tom: NeurIPS 2024 Workshop on Bayesian Decision-making and Uncertainty, strony: -), Wydawca: Online at OpenReview
    Data:
    konferencja 45640
    Status:
    Published
  3. Revisiting the Equivalence of Bayesian Neural Networks and Gaussian Processes: On the Importance of Learning Activations
    Authors:
    Marcin Sendera, Amin Sorkhei, Tomasz Kuśmierczyk
    Conference:
    Status:
    Accepted for publication