Research map / Continual learning / Continual fine-tuning of LLMs
Turning Back Without Forgetting: Selective Backward Refinement for Parameter-Efficient Continual Learning
Anushka Tiwari, Kaiyi Ji
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 task correlation selection (ablation) | SABER (T5-Large) · Long Sequence | AP: 77.59 -> 80.33 | Improved |
| add task correlation selection (ablation) | SABER (T5-Large) · Long Sequence | BWT: -1.78 -> 1.95 | Improved |
| add constrained backward refinement (ablation) | SABER (T5-Large) · Long Sequence | AP: 78.84 -> 80.33 | Improved |
| add constrained backward refinement (ablation) | SABER (T5-Large) · Long Sequence | BWT: -0.75 -> 1.95 | Improved |
| add task correlation selection (ablation) | SABER (T5-Large) · SuperNI | BWT: -0.84 -> 2.09 | Improved |
| add constrained backward refinement (ablation) | SABER (T5-Large) · SuperNI | AP: 41.33 -> 42.48 | Improved |
| add both components (ablation) | SABER (T5-Large) · Long Sequence | AP: 77.02 -> 80.33 | Improved |
| add task correlation selection (ablation) | SABER (LLaMA) · Long Sequence | AP: 81.26 -> 83.42 | Improved |
| add constrained backward refinement (ablation) | SABER (LLaMA) · Long Sequence | BWT: -0.87 -> 1.75 | Improved |
| add both components (ablation) | SABER (LLaMA) · SuperNI | AP: 45.12 -> 48.95 | Improved |
| orthogonal update instead of unconstrained backward update (ablation) | SABER backward prompt refinement · Long Sequence task pairs | avg Delta Acc: +0.007 -> +0.0215 | Improved |
| orthogonal update instead of same-subspace update (ablation) | SABER backward prompt refinement · Long Sequence task pairs | avg Delta Acc: -0.007 -> +0.0215 | Improved |
Closest papers by meaning
- A Unified and General Framework for Continual Learning (ICLR 2024 · similarity 0.89)
- Mixture of Experts Meets Prompt-Based Continual Learning (NeurIPS 2024 · similarity 0.89)
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.89)
- Advancing Prompt-Based Methods for Replay-Independent General Continual Learning (ICLR 2025 · similarity 0.89)
- InTAct: Interval-based Task Activation Consolidation for Continual Learning (NeurIPS 2026 · similarity 0.89)
- Mitigating Catastrophic Forgetting in Online Continual Learning by Modeling Previous Task Interrelations via Pareto Optimization (ICML 2024 · similarity 0.89)
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.
