Imperial College London
Portrait Picture

DR BERNHARD KAINZ

Faculty of Engineering, Department of Computing

Reader in Medical Image Computing
 
 
 
//

Contact

 

+44 (0)20 7594 8349b.kainz Website CV

 
 
//

Location

 

372Huxley BuildingSouth Kensington Campus

//

Summary

 

Publications

Citation

BibTex format

@inproceedings{Hinterreiter:2020:10.1007/978-3-030-61166-8_2,
author = {Hinterreiter, A and Streit, M and Kainz, B},
doi = {10.1007/978-3-030-61166-8_2},
pages = {13--22},
publisher = {Springer International Publishing},
title = {Projective latent interventions for understanding and fine-tuning classifiers},
url = {http://dx.doi.org/10.1007/978-3-030-61166-8_2},
year = {2020}
}
Download

RIS format (EndNote, RefMan)

TY  - CPAPER
AB  - High-dimensional latent representations learned by neural network classifiers are notoriously hard to interpret. Especially in medical applications, model developers and domain experts desire a better understanding of how these latent representations relate to the resulting classification performance. We present Projective Latent Interventions (PLIs), a technique for retraining classifiers by back-propagating manual changes made to low-dimensional embeddings of the latent space. The back-propagation is based on parametric approximations of   t -distributed stochastic neighbourhood embeddings. PLIs allow domain experts to control the latent decision space in an intuitive way in order to better match their expectations. For instance, the performance for specific pairs of classes can be enhanced by manually separating the class clusters in the embedding. We evaluate our technique on a real-world scenario in fetal ultrasound imaging.
AU  - Hinterreiter,A
AU  - Streit,M
AU  - Kainz,B
DO  - 10.1007/978-3-030-61166-8_2
EP  - 22
PB  - Springer International Publishing
PY  - 2020///
SN  - 0302-9743
SP  - 13
TI  - Projective latent interventions for understanding and fine-tuning classifiers
UR  - http://dx.doi.org/10.1007/978-3-030-61166-8_2
UR  - https://link.springer.com/chapter/10.1007%2F978-3-030-61166-8_2
UR  - http://hdl.handle.net/10044/1/83998
ER  -
Download