We introduce "Parallel Embeddings", a new technique that generalizes the classical Parallel Coordinates visualization technique to sequences of learned representations. This visualization technique is designed for concept-oriented "model comparison" tasks, allowing data scientists to understand qualitative differences in how models interpret input data. We compare user performance with our tool against Tensor Board Embedding Projector for understanding model accuracy and qualitative model differences. With our tool, users were more accurate and learned strategies for the tasks more quickly. Furthermore, users' analytical process in the comparison condition was positively influenced by using our tool beforehand.
Description
Parallel embeddings | Proceedings of the 25th International Conference on Intelligent User Interfaces
%0 Conference Paper
%1 Arendt_2020
%A Arendt, Dustin L.
%A Nur, Nasheen
%A Huang, Zhuanyi
%A Fair, Gabriel
%A Dou, Wenwen
%B Proceedings of the 25th International Conference on Intelligent User Interfaces
%D 2020
%I ACM
%K deep-learning explanation information-visualization interactive-machine-learning iui2020 transparency
%P 259-274
%R 10.1145/3377325.3377514
%T Parallel embeddings: A visualization technique for contrasting learned representations
%U https://doi.org/10.1145%2F3377325.3377514
%X We introduce "Parallel Embeddings", a new technique that generalizes the classical Parallel Coordinates visualization technique to sequences of learned representations. This visualization technique is designed for concept-oriented "model comparison" tasks, allowing data scientists to understand qualitative differences in how models interpret input data. We compare user performance with our tool against Tensor Board Embedding Projector for understanding model accuracy and qualitative model differences. With our tool, users were more accurate and learned strategies for the tasks more quickly. Furthermore, users' analytical process in the comparison condition was positively influenced by using our tool beforehand.
@inproceedings{Arendt_2020,
abstract = {We introduce "Parallel Embeddings", a new technique that generalizes the classical Parallel Coordinates visualization technique to sequences of learned representations. This visualization technique is designed for concept-oriented "model comparison" tasks, allowing data scientists to understand qualitative differences in how models interpret input data. We compare user performance with our tool against Tensor Board Embedding Projector for understanding model accuracy and qualitative model differences. With our tool, users were more accurate and learned strategies for the tasks more quickly. Furthermore, users' analytical process in the comparison condition was positively influenced by using our tool beforehand.
},
added-at = {2020-08-02T22:47:36.000+0200},
author = {Arendt, Dustin L. and Nur, Nasheen and Huang, Zhuanyi and Fair, Gabriel and Dou, Wenwen},
biburl = {https://www.bibsonomy.org/bibtex/2cf2ece2b7da2928a972b4ec89cc1244e/brusilovsky},
booktitle = {Proceedings of the 25th International Conference on Intelligent User Interfaces},
description = {Parallel embeddings | Proceedings of the 25th International Conference on Intelligent User Interfaces},
doi = {10.1145/3377325.3377514},
interhash = {4b7c587f25e9453065fe3618e7093856},
intrahash = {cf2ece2b7da2928a972b4ec89cc1244e},
keywords = {deep-learning explanation information-visualization interactive-machine-learning iui2020 transparency},
month = mar,
pages = {259-274},
publisher = {{ACM}},
timestamp = {2020-12-07T21:15:38.000+0100},
title = {Parallel embeddings: A visualization technique for contrasting learned representations},
url = {https://doi.org/10.1145%2F3377325.3377514},
year = 2020
}