Research map / Architectures and efficiency / Pruning and efficient tuning
AD-DROP: Attribution-Driven Dropout for Robust Language Model Fine-Tuning
Tao Yang, Jinghao Deng, Xiaojun Quan, Qifan Wang, Shaoliang Nie
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 |
|---|---|---|---|
| Add AD-Drop to BERT_base fine-tuning | BERT_base · GLUE development | Average: 83.23 -> 85.21 | Improved |
| Add AD-Drop to RoBERTa_base fine-tuning | RoBERTa_base · GLUE development | Average: 86.69 -> 87.98 | Improved |
| Add AD-Drop to BERT_base fine-tuning | BERT_base · GLUE test | Average: 80.78 -> 81.65 | Improved |
| Add AD-Drop to RoBERTa_base fine-tuning | RoBERTa_base · GLUE test | Average: 84.14 -> 84.76 | Improved |
| Add AD-Drop during fine-tuning | RoBERTa_base · HANS validation | Accuracy: 69.83 -> 70.49 | Improved |
| Add AD-Drop during fine-tuning | RoBERTa_base · PAWS-X test | Accuracy: 47.90 -> 51.25 | Improved |
| Replace gradient attribution with integrated gradient attribution (ablation) | BERT_base + AD-Drop · RTE development | Accuracy: 75.1 -> 74.4 | Hurt |
| Replace gradient attribution with attention weights (ablation) | BERT_base + AD-Drop · RTE development | Accuracy: 75.1 -> 74.7 | Hurt |
| Replace gradient-based discard region with random discard region (ablation) | BERT_base + AD-Drop · RTE development | Accuracy: 75.1 -> 74.7 | Hurt |
| Use gold labels instead of pseudo labels for attribution (ablation) | BERT_base + AD-Drop · RTE development | Accuracy: 75.1 -> 74.4 | Hurt |
| Add cross-tuning, alternating original fine-tuning and AD-Drop (ablation) | BERT_base + AD-Drop · RTE development | Accuracy: 71.5 -> 75.1 | Improved |
| Replace gradient attribution with integrated gradient attribution (ablation) | RoBERTa_base + AD-Drop · RTE development | Accuracy: 84.1 -> 82.7 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ELECTRA_base — ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators (ICLR 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DropBP: Accelerating Fine-Tuning of Large Language Models by Dropping Backward Propagation (NeurIPS 2024 · similarity 0.87)
- Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models (ICML 2024 · similarity 0.87)
- An Emulator for Fine-tuning Large Language Models using Small Language Models (ICLR 2024 · similarity 0.86)
- When Attention Sink Emerges in Language Models: An Empirical View (ICLR 2025 · similarity 0.86)
- Unveiling and Harnessing Hidden Attention Sinks: Enhancing Large Language Models without Training through Attention Calibration (ICML 2024 · similarity 0.86)
- Helpful or Harmful Data? Fine-tuning-free Shapley Attribution for Explaining Language Model Predictions (ICML 2024 · similarity 0.86)
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.
