Research map / Reinforcement learning / Visual and representation RL
Learning Data Manipulation for Augmentation and Weighting
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| fine-tune the augmentation LM jointly with the classifier instead of keeping it fixed (ablation) | BERT text augmentation · SST-5 | accuracy: 34.84 -> 37.03 | Helped |
| fine-tune the augmentation LM jointly instead of keeping it fixed (ablation) | BERT text augmentation · IMDB | accuracy: 63.65 -> 65.62 | Helped |
| fine-tune the augmentation LM jointly instead of keeping it fixed (ablation) | BERT text augmentation · TREC | accuracy: 88.28 -> 89.15 | Helped |
| use synonym augmentation instead of no augmentation | BERT · SST-5 | accuracy: 33.32 -> 32.45 | Hurt |
| treat data weights as persistent learned parameters instead of re-estimating each step | data weighting · SST-5 | accuracy: 36.09 -> 36.51 | Helped |
| use learned data weighting instead of base model | BERT · Imbalanced SST-2 20:1000 | accuracy: 54.91 -> 75.08 | Helped |
| use learned data weighting instead of base model | ResNet-34 · Imbalanced CIFAR10 20:1000 | accuracy: 72.20 -> 75.32 | Helped |
| use inverse-class-frequency proportion weighting | ResNet-34 · Imbalanced CIFAR10 100:1000 | accuracy: 86.42 -> 84.26 | Hurt |
| use learned data weighting with ImageNet-pretrained init vs random init (ablation) | ResNet-34 · CIFAR10 subset | accuracy: 22.98 -> 24.92 | Helped |
| apply LM-based text augmentation to class-imbalanced data | BERT SST-2 imbalanced · Imbalanced SST-2 | accuracy: around 50% | Hurt |
| train base model on merged train+validation data | BERT · Imbalanced SST-2 50:1000 | accuracy: 67.73 -> 55.90 | Hurt |
Closest papers by meaning
- Data Augmentation as Feature Manipulation (ICML 2022 · similarity 0.87)
- Revisiting Data Augmentation in Deep Reinforcement Learning (ICLR 2024 · similarity 0.85)
- Understanding when Dynamics-Invariant Data Augmentations Benefit Model-free Reinforcement Learning Updates (ICLR 2024 · similarity 0.85)
- Data Augmentation with Diffusion for Open-Set Semi-Supervised Learning (NeurIPS 2024 · similarity 0.84)
- Effective Data Augmentation With Diffusion Models (ICLR 2024 · similarity 0.84)
- Data Attribution for Text-to-Image Models by Unlearning Synthesized Images (NeurIPS 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.