Robustness of Deep Learning in Image Reconstruction
Vegard Antun1
1University of Oslo, Oslo, Norway

Synopsis

We all know that artificial intelligence (AI) can do marvelous things and that it is currently being incorporated into many industries. Yet, modern AI has an Achilles heel: It seems universally non-robust. That is, it can be unstable to tiny perturbations or generalize in unpredictable ways, both of which can lead to AI generated hallucinations. In this tutorial, we will investigate the many reasons for instabilities, unpredictable generalization and hallucinations in AI based image reconstruction. Furthermore, we will provide guidance on how to reduce the issue of non-robustness and untrustworthy outputs in AI based image reconstruction.

Acknowledgements

No acknowledgement found.

References

No reference found.
Proc. Intl. Soc. Mag. Reson. Med. 30 (2022)