GeomCA: Geometric Evaluation of Data Representations

Petra Poklukar, Anastasiia Varava, Danica Kragic
Proceedings of the 38th International Conference on Machine Learning, PMLR 139:8588-8598, 2021.

Abstract

Evaluating the quality of learned representations without relying on a downstream task remains one of the challenges in representation learning. In this work, we present Geometric Component Analysis (GeomCA) algorithm that evaluates representation spaces based on their geometric and topological properties. GeomCA can be applied to representations of any dimension, independently of the model that generated them. We demonstrate its applicability by analyzing representations obtained from a variety of scenarios, such as contrastive learning models, generative models and supervised learning models.

Cite this Paper

BibTeX
@InProceedings{pmlr-v139-poklukar21a, title = {GeomCA: Geometric Evaluation of Data Representations}, author = {Poklukar, Petra and Varava, Anastasiia and Kragic, Danica}, booktitle = {Proceedings of the 38th International Conference on Machine Learning}, pages = {8588--8598}, year = {2021}, editor = {Meila, Marina and Zhang, Tong}, volume = {139}, series = {Proceedings of Machine Learning Research}, month = {18--24 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v139/poklukar21a/poklukar21a.pdf}, url = {https://proceedings.mlr.press/v139/poklukar21a.html}, abstract = {Evaluating the quality of learned representations without relying on a downstream task remains one of the challenges in representation learning. In this work, we present Geometric Component Analysis (GeomCA) algorithm that evaluates representation spaces based on their geometric and topological properties. GeomCA can be applied to representations of any dimension, independently of the model that generated them. We demonstrate its applicability by analyzing representations obtained from a variety of scenarios, such as contrastive learning models, generative models and supervised learning models.} }
Endnote
%0 Conference Paper %T GeomCA: Geometric Evaluation of Data Representations %A Petra Poklukar %A Anastasiia Varava %A Danica Kragic %B Proceedings of the 38th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2021 %E Marina Meila %E Tong Zhang %F pmlr-v139-poklukar21a %I PMLR %P 8588--8598 %U https://proceedings.mlr.press/v139/poklukar21a.html %V 139 %X Evaluating the quality of learned representations without relying on a downstream task remains one of the challenges in representation learning. In this work, we present Geometric Component Analysis (GeomCA) algorithm that evaluates representation spaces based on their geometric and topological properties. GeomCA can be applied to representations of any dimension, independently of the model that generated them. We demonstrate its applicability by analyzing representations obtained from a variety of scenarios, such as contrastive learning models, generative models and supervised learning models.
APA
Poklukar, P., Varava, A. & Kragic, D.. (2021). GeomCA: Geometric Evaluation of Data Representations. Proceedings of the 38th International Conference on Machine Learning, in Proceedings of Machine Learning Research 139:8588-8598 Available from https://proceedings.mlr.press/v139/poklukar21a.html.

Related Material