Towards robust interpretability with self-explaining neural networks
© 2018 Curran Associates Inc.All rights reserved. Most recent work on interpretability of complex machine learning models has focused on estimating a posteriori explanations for previously trained models around specific predictions. Self-explaining models where interpretability plays a key role alre...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
2021
|
| Online Access: | https://hdl.handle.net/1721.1/137669.3 |
| _version_ | 1811079988278984704 |
|---|---|
| author | Jaakkola, Tommi Alvarez Melis, David |
| author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
| author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Jaakkola, Tommi Alvarez Melis, David |
| author_sort | Jaakkola, Tommi |
| collection | MIT |
| description | © 2018 Curran Associates Inc.All rights reserved. Most recent work on interpretability of complex machine learning models has focused on estimating a posteriori explanations for previously trained models around specific predictions. Self-explaining models where interpretability plays a key role already during learning have received much less attention. We propose three desiderata for explanations in general - explicitness, faithfulness, and stability - and show that existing methods do not satisfy them. In response, we design self-explaining models in stages, progressively generalizing linear classifiers to complex yet architecturally explicit models. Faithfulness and stability are enforced via regularization specifically tailored to such models. Experimental results across various benchmark datasets show that our framework offers a promising direction for reconciling model complexity and interpretability. |
| first_indexed | 2024-09-23T11:23:42Z |
| format | Article |
| id | mit-1721.1/137669.3 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T11:23:42Z |
| publishDate | 2021 |
| record_format | dspace |
| spelling | mit-1721.1/137669.32021-12-21T20:30:27Z Towards robust interpretability with self-explaining neural networks Jaakkola, Tommi Alvarez Melis, David Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory © 2018 Curran Associates Inc.All rights reserved. Most recent work on interpretability of complex machine learning models has focused on estimating a posteriori explanations for previously trained models around specific predictions. Self-explaining models where interpretability plays a key role already during learning have received much less attention. We propose three desiderata for explanations in general - explicitness, faithfulness, and stability - and show that existing methods do not satisfy them. In response, we design self-explaining models in stages, progressively generalizing linear classifiers to complex yet architecturally explicit models. Faithfulness and stability are enforced via regularization specifically tailored to such models. Experimental results across various benchmark datasets show that our framework offers a promising direction for reconciling model complexity and interpretability. 2021-12-21T20:30:26Z 2021-11-08T14:41:45Z 2021-12-21T20:30:26Z 2018 2019-05-31T16:23:58Z Article http://purl.org/eprint/type/ConferencePaper https://hdl.handle.net/1721.1/137669.3 Jaakkola, Tommi and Alvarez Melis, David. 2018. "Towards robust interpretability with self-explaining neural networks." en https://papers.nips.cc/paper/8003-towards-robust-interpretability-with-self-explaining-neural-networks Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use. application/octet-stream Neural Information Processing Systems (NIPS) |
| spellingShingle | Jaakkola, Tommi Alvarez Melis, David Towards robust interpretability with self-explaining neural networks |
| title | Towards robust interpretability with self-explaining neural networks |
| title_full | Towards robust interpretability with self-explaining neural networks |
| title_fullStr | Towards robust interpretability with self-explaining neural networks |
| title_full_unstemmed | Towards robust interpretability with self-explaining neural networks |
| title_short | Towards robust interpretability with self-explaining neural networks |
| title_sort | towards robust interpretability with self explaining neural networks |
| url | https://hdl.handle.net/1721.1/137669.3 |
| work_keys_str_mv | AT jaakkolatommi towardsrobustinterpretabilitywithselfexplainingneuralnetworks AT alvarezmelisdavid towardsrobustinterpretabilitywithselfexplainingneuralnetworks |