Research map / Label-efficient and robust learning / Recognition and long-tail
Accessing Higher Dimensions for Unsupervised Word Translation
Sida Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use co-occurrence association matrix (coocmap) instead of fasttext vectors (ablation) | vecmap · enwiki-eswiki | data to start working (MB): 70 -> 9 | Improved |
| coocmap instead of fasttext vectors (ablation) | vecmap · enwiki-zhwiki | data to start working (MB): 700 -> 14 | Improved |
| add clip and drop to coocmap (ablation) | coocmap · enwiki-esnews | data to reach 50% accuracy (MB): 100 -> 40 | Improved |
| add clip and drop to coocmap (ablation) | coocmap · enwiki-hunews | precision@1: 5% -> 50% | Improved |
| add clip (and drop) on ennews-zhwiki (ablation) | coocmap · ennews-zhwiki | accuracy | Improved |
| reduce fasttext dimension from 300 to 100 (ablation) | vecmap-fasttext · enparl-huparl | data to start working (MB): 8 -> 50 | Improved |
| add clip to fasttext-based coocmap (ablation) | coocmap-fasttext · enparl-huparl | accuracy | Improved |
| increase clip percentile from 1% to 1.5% (ablation) | coocmap-clip · ennews-zhwiki | accuracy | No clear effect |
| truncate tail of SVD to 2000 of 5000 dimensions (ablation) | coocmap-clip-drop · ennews-zhwiki 300MB | accuracy | Improved |
| remove CSLS hubness mitigation (plain bidirectional matching) (ablation) | coocmap · enwiki-eswiki | accuracy | Hurt |
| turn off fasttext subword information (ablation) | fasttext · word translation | accuracy | No clear effect |
| change initialization (full coocmap init vs own init) (ablation) | vecmap-fasttext · enwiki-esnews | accuracy | No clear effect |
Closest papers by meaning
- Comparing Unsupervised Word Translation Methods Step by Step (NeurIPS 2019 · similarity 0.86)
- Visual Grounding in Video for Unsupervised Word Translation (CVPR 2020 · similarity 0.86)
- Unsupervised Machine Translation Using Monolingual Corpora Only (ICLR 2018 · similarity 0.86)
- Unsupervised Neural Machine Translation (ICLR 2018 · similarity 0.86)
- Offline bilingual word vectors, orthogonal transformations and the inverted softmax (ICLR 2017 · similarity 0.85)
- Rethinking Channel Dimensions to Isolate Outliers for Low-bit Weight Quantization of Large Language Models (ICLR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
