TY - GEN
T1 - Projection-Wise Disentangling for Fair and Interpretable Representation Learning
T2 - 24th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2021
AU - Liu, Xianjing
AU - Li, Bo
AU - Bron, Esther E.
AU - Niessen, Wiro J.
AU - Wolvius, Eppo B.
AU - Roshchupkin, Gennady V.
PY - 2021
Y1 - 2021
N2 - Confounding bias is a crucial problem when applying machine learning to practice, especially in clinical practice. We consider the problem of learning representations independent to multiple biases. In literature, this is mostly solved by purging the bias information from learned representations. We however expect this strategy to harm the diversity of information in the representation, and thus limiting its prospective usage (e.g., interpretation). Therefore, we propose to mitigate the bias while keeping almost all information in the latent representations, which enables us to observe and interpret them as well. To achieve this, we project latent features onto a learned vector direction, and enforce the independence between biases and projected features rather than all learned features. To interpret the mapping between projected features and input data, we propose projection-wise disentangling: a sampling and reconstruction along the learned vector direction. The proposed method was evaluated on the analysis of 3D facial shape and patient characteristics (N = 5011). Experiments showed that this conceptually simple method achieved state-of-the-art fair prediction performance and interpretability, showing its great potential for clinical applications.
AB - Confounding bias is a crucial problem when applying machine learning to practice, especially in clinical practice. We consider the problem of learning representations independent to multiple biases. In literature, this is mostly solved by purging the bias information from learned representations. We however expect this strategy to harm the diversity of information in the representation, and thus limiting its prospective usage (e.g., interpretation). Therefore, we propose to mitigate the bias while keeping almost all information in the latent representations, which enables us to observe and interpret them as well. To achieve this, we project latent features onto a learned vector direction, and enforce the independence between biases and projected features rather than all learned features. To interpret the mapping between projected features and input data, we propose projection-wise disentangling: a sampling and reconstruction along the learned vector direction. The proposed method was evaluated on the analysis of 3D facial shape and patient characteristics (N = 5011). Experiments showed that this conceptually simple method achieved state-of-the-art fair prediction performance and interpretability, showing its great potential for clinical applications.
KW - 3D shape analysis
KW - Disentangled representation learning
KW - Fair representation learning
KW - Interpretability
UR - http://www.scopus.com/inward/record.url?scp=85116399819&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-87240-3_78
DO - 10.1007/978-3-030-87240-3_78
M3 - Conference contribution
AN - SCOPUS:85116399819
SN - 9783030872397
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 814
EP - 823
BT - Medical Image Computing and Computer Assisted Intervention – MICCAI 2021 - 24th International Conference, Proceedings
A2 - de Bruijne, Marleen
A2 - Cattin, Philippe C.
A2 - Cotin, Stéphane
A2 - Padoy, Nicolas
A2 - Speidel, Stefanie
A2 - Zheng, Yefeng
A2 - Essert, Caroline
PB - Springer
Y2 - 27 September 2021 through 1 October 2021
ER -