Efficient Contrastive Explanations on Demand The main purpose of this CoRR report is to serve as a timestamp against the copycats.
Yacine Izza, Joao Marques-Silva · 2024
Read Efficient Contrastive Explanations on Demand The main purpose of this CoRR report is to serve as a timestamp against the copycats. by Yacine Izza, Joao Marques-Silva (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-18262.txt.
Description
Yacine Izza and Joao Marques-Silva speed up contrastive explanations in a 2024 report. Such an explanation says which features would have to change to flip what a model decides. Formal forms of them tie closely to adversarial robustness. The hard part was working them out for models with very large feature counts.
Their new algorithms use checks of that robustness to cut down the search. The method can also list the explanations and find the smallest ones. Tests show clear gains in speed over past methods. Deep networks with thousands of features are now in reach. The authors note that the report also serves as a timestamped claim of priority. These explanations matter for auditing automated decisions, and faster work makes such checks a realistic routine.
Catalog record
| CREATOR | Yacine Izza, Joao Marques-Silva |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-18262