Research map / Continual learning / Online and replay-based
RATT: Recurrent Attention to Transient Tasks for Continual Image Captioning
Riccardo Del Chiaro, Bartłomiej Twardowski, Andrew D. Bagdanov, Joost van de Weijer
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 RATT attention masks instead of fine-tuning | LSTM captioner · MS-COCO animals task | BLEU-4: .0816 -> .2468 | Improved |
| use RATT instead of EWC | LSTM captioner · MS-COCO sports task | BLEU-4: .2182 -> .3161 | Improved |
| use RATT instead of fine-tuning | LSTM captioner · MS-COCO interior (last) task | BLEU-4: .2712 -> .2727 | No clear effect |
| use RATT instead of fine-tuning | LSTM captioner · MS-COCO transport task | % forgetting: 59.1 -> 0.0 | Improved |
| use RATT instead of LwF | LSTM captioner · Flickr30k instruments (last) task | BLEU-4: .1862 -> .2386 | Improved |
| use RATT instead of fine-tuning | LSTM captioner · Flickr30k scene task | BLEU-4: .1074 -> .1548 | Improved |
| use RATT instead of LwF (METEOR/CIDEr mixed on first Flickr task) | LSTM captioner · Flickr30k scene task | METEOR: .1851 -> .1710 | Hurt |
| add hidden-state attention mask only (progressive ablation) (ablation) | fine-tuning captioner · MS-COCO | BLEU-4 forgetting | Mixed |
| add embedding masks and combine all masks (ablation) | RATT · MS-COCO | BLEU-4 forgetting | Improved |
| increase s_max (sharper masks) (ablation) | RATT · MS-COCO | old-task BLEU-4 | Mixed |
Closest papers by meaning
- A continual learning survey: Defying forgetting in classification tasks (TPAMI 2021 · similarity 0.90)
- CaPCL: Caption-Preserved Continual Learning for Text-to-Image Retrieval (ECCV 2026 · similarity 0.89)
- Enhancing Visual Continual Learning with Language-Guided Supervision (CVPR 2024 · similarity 0.89)
- CRNet: A Fast Continual Learning Framework With Random Theory (TPAMI 2023 · similarity 0.89)
- Convolutional Prompting meets Language Models for Continual Learning (CVPR 2024 · similarity 0.88)
- Anytime Continual Learning for Open Vocabulary Classification (ECCV 2024 · similarity 0.88)
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.
