Just shy over a year ago, the Quantus toolkit v0.1.1 has been shared with the Machine Learning (ML) community as a pre-print on arXiv.org.
It can be found here. Quantus is a toolbox for reproducible Explainable Artificial Intelligence (XAI)(-based) evaluations of Artificial Intelligence (AI). Since its initial release, Quantus has seen some considerable growth, leading to (among other things) a recent peer-reviewed publication in the prestigious Journal of Machine Learning Research (JMLR, the MLOSS track) and a release version v0.3.4 ! The paper is available under Open Acess license at https://www.jmlr.org/papers/v24/22-0142.html.
In Summary, Quantus collects and provides 30+ metrics for XAI evaluation from 6 categories as proposed in related literature and supports different data types (image, time-series, tabular, NLP next up!) and models (PyTorch, TensorFlow), via an easy-to-use and extend API and code structure. Go check out the best versioin of Quantus there is today on its githhub project page at https://github.com/understandable-machine-intelligence-lab/Quantus.
Fig. 1 An illustration on how the Quantus toolkit can be used to derive a variety of measures from XAI in order to assess the explanations of AI behaviour themselves, or the underlying model.
The Evolution of Quantus
Significant upgrades of the Quantus package since its initial release on February 11, 2022 include:
-
New metrics were added to the Robustness and Faithfulness Categories.
-
Extended built-in support for explanation methods via Captum (https://captum.ai/) and tf-explain (https://github.com/sicara/tf-explain), next to the option to process pre-computed attributions from custom and other XAI methods.
-
Considerable Code optimizations have been added, to (1) speed up and parallelize computations (see API references [1] and [2], and (2) make it easier for developers to add new code (see [3]).
-
The documentation has been improved to provide clarity and quality of life for users of the package (see [4]).
Building a Community
Over the last year, a healthy and striving community has developed around Quantus, which reflects in:
-
One community member (D. Krakowczyk from Potsdam University) joining our ranks as a Co-Author on the Quantus JMLR paper!
-
Over 300 stars being awarded to Quantus’ GitHub repository [5] as of today, with
-
Over 50 forks being created from the project and
-
Over 231 pull requests from 14 contributors committed to date.
-
According to https://pepy.tech/project/quantus, we count over 13,500 pip-installs to the present date, 𝘯𝘰𝘵 counting any direct clones or downloads from GitHub or similar installs!
Quantus in Action
Quantus can already be found in a diverse set of applications in the XAI and ML communities, including the evaluation of explanations for machine learning models in Climate Science [6] and Healthcare (content-based image retrieval of CT liver images [7], Eye Tracking applications (published at NeurIPS) [8], or for evaluating explanations for the development of human-machine interfaces for prosthetic hand control (published in IEEE) [9]), and in (X)AI itself as a tool for hyperparameter optimization for XAI methods [10]. Together, the journal paper and pre-print have been attributed with 17 citations in literature as of today.
Acknowledgements
The work on Quantus was partly funded by the German Ministry for Education and Research through project Explaining 4.0 (ref. 01IS20055) and BIFOLD (ref. 01IS18025A and ref. 01IS18037A), the Investitionsbank Berlin through project BerDiBa (grant no. 10174498), as well as the European Union’s Horizon 2020 programme through iToBoS (grant no. 965221).
Sebastian Lapuschkin, Fraunhofer Heinrich-Hertz-Institute
[1] https://quantus.readthedocs.io/en/latest/docs_api/quantus.metrics.base_batched.html
[2] https://github.com/understandable-machine-intelligence-lab/Quantus/releases/tag/v0.3.0
[3] https://github.com/understandable-machine-intelligence-lab/Quantus/releases/tag/v0.2.0
[4] https://quantus.readthedocs.io/en/latest/
[5] https://github.com/understandable-machine-intelligence-lab/Quantus
[6] https://arxiv.org/abs/2303.00652
[7] https://arxiv.org/abs/2207.04812
[8] https://openreview.net/forum?id=GOLdDAP2AtI
[9] https://ieeexplore.ieee.org/abstract/document/9999246
[10] https://arxiv.org/abs/2211.17174