Research map / Continual learning / Domain and prompt-based
Progressive Prompts: Continual Learning for Language Models
Anastasia Razdaibiedina, Yuning Mao, Rui Hou, Madian Khabsa, Mike Lewis, Amjad Almahairi
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 |
|---|---|---|---|
| progressively concatenate a new task prompt with frozen prompts from earlier tasks | T5-large · T5 CL benchmark, average across three task orders | accuracy: 52.7 -> 75.1 | Improved |
| progressively concatenate a new task prompt with frozen prompts from earlier tasks | BERT-base · BERT CL benchmark, average across four task orders | accuracy: 76.3 -> 77.9 | Improved |
| use progressive prompts instead of LFPT5 | T5-Large · 15-task CL, 20 samples/class | average test performance: 54.3 -> 76.2 | Improved |
| use progressive prompts instead of IDBR | BERT-base · 15-task CL, 20 samples/class | average test performance: 36.8 -> 53.5 | Improved |
| transfer a learned source-task prompt to a related target task by concatenation | T5-Large · IMDb → SST2, all training data | performance: 93.6 -> 95.8 | Improved |
| transfer a learned source-task prompt to a related target task by concatenation | T5-Large · MRPC → QQP, all training data | performance: 91.1 -> 90.9 | Hurt |
| transfer a learned source-task prompt to a related target task by concatenation | T5-Large · six related task pairs, 5 samples/class | average performance: 57.5 -> 68.2 | Mixed |
| add GLUE prompts before task-specific SuperGLUE prompts | T5 · SuperGLUE validation | SuperGLUE score: 74.5 -> 77.2 | Improved |
| add residual MLP reparameterization to prompt tuning (ablation) | BERT-base · IMDb, prompt length 5 | validation performance: 85.8_{0.7} -> 91.4_{0.8} | Improved |
| use a plain MLP reparameterization without the residual connection (ablation) | BERT-base prompt tuning · IMDb, prompt length 5 | validation performance: 85.8_{0.7} -> 83.4_{0.9} | Hurt |
| add residual MLP reparameterization to prompt tuning (ablation) | BERT-base · QQP, prompt length 5 | validation performance: 72.9_{0.8} -> 76.6_{1.2} | Improved |
| add residual MLP reparameterization to prompt tuning (ablation) | BERT-base · RTE, prompt length 5 | validation performance: 65.0_{2.3} -> 65.0_{3.5} | No clear effect |
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.
- LFPT5 — LFPT5: A Unified Framework for Lifelong Few-shot Language Learning Based on Prompt Tuning of T5 (ICLR 2022) · better on 4 of 4 shared results
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
- Convolutional Prompting meets Language Models for Continual Learning (CVPR 2024 · similarity 0.90)
- Learn more, but bother less: parameter efficient continual learning (NeurIPS 2024 · similarity 0.90)
- Scalable Language Model with Generalized Continual Learning (ICLR 2024 · similarity 0.89)
- Spurious Forgetting in Continual Learning of Language Models (ICLR 2025 · similarity 0.89)
- Adaptive Prompting for Continual Relation Extraction: A Within-Task Variance Perspective (AAAI 2025 · similarity 0.88)
- One Size Fits All for Semantic Shifts: Adaptive Prompt Tuning for Continual Learning (ICML 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.
