Research map / Architectures and efficiency / RNNs and normalization
Robust Optimization for Multilingual Translation with Imbalanced Data
Xian Li, Hongyu Gong
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 curvature-aware learned task weights instead of T=1 sampling | Multilingual translation · TED related | average BLEU (All): 20.0 -> 20.7 | Improved |
| Use curvature-aware learned task weights instead of T=1 sampling | Multilingual translation · TED diverse | average BLEU (Lo): 15.7 -> 16.1 | Improved |
| Use curvature-aware learned task weights instead of T=5 sampling | Multilingual translation · WMT | average BLEU (Lo): +2.2 BLEU | Improved |
| Use curvature-aware learned task weights instead of T=5 sampling | Multilingual translation · WMT | average BLEU (All): 20.9 -> 23.5 | Improved |
| Use curvature-aware learned task weights instead of T=5 sampling | Multilingual translation · OPUS-100 | average BLEU (Lo): +1.3 BLEU | Improved |
| Remove LayerNorm parameters (ablation) | TED diverse translation with T=1 baseline · TED diverse | average BLEU (Lo): 22.8 -> 23.1 | Improved |
| Add CATS while removing LayerNorm parameters (ablation) | TED diverse translation · TED diverse | average BLEU (Lo): 23.1 -> 23.6 | Improved |
| Upsample low-resource data with T=5 rather than proportional sampling | Multilingual translation · TED related | average BLEU (All): 20.0 -> 19.9 | No clear effect |
| Upsample low-resource data with T=5 rather than proportional sampling | Multilingual translation · TED diverse | average BLEU (All): 23.5 -> 22.5 | Hurt |
Closest papers by meaning
- Exploring Polyglot Harmony: On Multilingual Data Allocation for Large Language Models Pretraining (NeurIPS 2025 · similarity 0.90)
- X-ALMA: Plug & Play Modules and Adaptive Rejection for Quality Translation at Scale (ICLR 2025 · similarity 0.88)
- Breaking Physical and Linguistic Borders: Multilingual Federated Prompt Tuning for Low-Resource Languages (ICLR 2024 · similarity 0.88)
- Mufu: Multilingual Fused Learning for Low-Resource Translation with LLM (ICLR 2025 · similarity 0.88)
- When Scaling Meets LLM Finetuning: The Effect of Data, Model and Finetuning Method (ICLR 2024 · similarity 0.87)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.87)
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.
