Mitigating Biases for Instruction-following Language Models via Bias Neurons Elimination
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2023
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866916274080055296 |
|---|---|
| author | Yang, Nakyeong Kang, Taegwan Choi, Jungkyu Lee, Honglak Jung, Kyomin |
| author_facet | Yang, Nakyeong Kang, Taegwan Choi, Jungkyu Lee, Honglak Jung, Kyomin |
| contents | Instruction-following language models often show undesirable biases. These undesirable biases may be accelerated in the real-world usage of language models, where a wide range of instructions is used through zero-shot example prompting. To solve this problem, we first define the bias neuron, which significantly affects biased outputs, and prove its existence empirically. Furthermore, we propose a novel and practical bias mitigation method, CRISPR, to eliminate bias neurons of language models in instruction-following settings. CRISPR automatically determines biased outputs and categorizes neurons that affect the biased outputs as bias neurons using an explainability method. Experimental results demonstrate the effectiveness of our method in mitigating biases under zero-shot instruction-following settings without losing the model's task performance and existing knowledge. The experimental results reveal the generalizability of our method as it shows robustness under various instructions and datasets. Surprisingly, our method can mitigate the bias in language models by eliminating only a few neurons (at least three). |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2311_09627 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Mitigating Biases for Instruction-following Language Models via Bias Neurons Elimination Yang, Nakyeong Kang, Taegwan Choi, Jungkyu Lee, Honglak Jung, Kyomin Artificial Intelligence Computation and Language Machine Learning Instruction-following language models often show undesirable biases. These undesirable biases may be accelerated in the real-world usage of language models, where a wide range of instructions is used through zero-shot example prompting. To solve this problem, we first define the bias neuron, which significantly affects biased outputs, and prove its existence empirically. Furthermore, we propose a novel and practical bias mitigation method, CRISPR, to eliminate bias neurons of language models in instruction-following settings. CRISPR automatically determines biased outputs and categorizes neurons that affect the biased outputs as bias neurons using an explainability method. Experimental results demonstrate the effectiveness of our method in mitigating biases under zero-shot instruction-following settings without losing the model's task performance and existing knowledge. The experimental results reveal the generalizability of our method as it shows robustness under various instructions and datasets. Surprisingly, our method can mitigate the bias in language models by eliminating only a few neurons (at least three). |
| title | Mitigating Biases for Instruction-following Language Models via Bias Neurons Elimination |
| topic | Artificial Intelligence Computation and Language Machine Learning |
| url | https://arxiv.org/abs/2311.09627 |