Improving LLM Attributions with Randomized Path-Integration

Oren Barkan, Yehonatan Elisha, Yonatan Toib, Jonathan Weill, Noam Koenigstein


Abstract
We present Randomized Path-Integration (RPI) - a path-integration method for explaining language models via randomization of the integration path over the attention information in the model. RPI employs integration on internal attention scores and their gradients along a randomized path, which is dynamically established between a baseline representation and the attention scores of the model. The inherent randomness in the integration path originates from modeling the baseline representation as a randomly drawn tensor from a Gaussian diffusion process. As a consequence, RPI generates diverse baselines, yielding a set of candidate attribution maps. This set facilitates the selection of the most effective attribution map based on the specific metric at hand. We present an extensive evaluation, encompassing 11 explanation methods and 5 language models, including the Llama2 and Mistral models. Our results demonstrate that RPI outperforms latest state-of-the-art methods across 4 datasets and 5 evaluation metrics.
Anthology ID:
2024.findings-emnlp.551
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2024
Month:
November
Year:
2024
Address:
Miami, Florida, USA
Editors:
Yaser Al-Onaizan, Mohit Bansal, Yun-Nung Chen
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
9430–9446
Language:
URL:
https://aclanthology.org/2024.findings-emnlp.551
DOI:
Bibkey:
Cite (ACL):
Oren Barkan, Yehonatan Elisha, Yonatan Toib, Jonathan Weill, and Noam Koenigstein. 2024. Improving LLM Attributions with Randomized Path-Integration. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 9430–9446, Miami, Florida, USA. Association for Computational Linguistics.
Cite (Informal):
Improving LLM Attributions with Randomized Path-Integration (Barkan et al., Findings 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.findings-emnlp.551.pdf