Debiasing performance on GloVe word embeddings. FaRM significantly outperforms INLP (Ravfogel et al., 2020) in guarding gender information. Best debiasing results are in bold.
| Method . | Accuracy (↓) . | MDL (↑) . | Rank (↑) . |
|---|---|---|---|
| GloVe | 100.0 | 0.1 | 300 |
| INLP | 86.3 | 8.6 | 210 |
| FaRM | 53.9 | 24.6 | 247 |
| Method . | Accuracy (↓) . | MDL (↑) . | Rank (↑) . |
|---|---|---|---|
| GloVe | 100.0 | 0.1 | 300 |
| INLP | 86.3 | 8.6 | 210 |
| FaRM | 53.9 | 24.6 | 247 |