Research map / Continual learning / Continual fine-tuning of LLMs
Memory-based Parameter Adaptation
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 |
|---|---|---|---|
| Use MbPA instead of a plain parametric model | ResNetV1 incremental ImageNet classification · ImageNet novel classes | Top 1 accuracy: 46.2 % vs 16.2 % | Helped |
| Use MbPA instead of a memory-based mixture | ResNetV1 incremental ImageNet classification · ImageNet novel classes | Top 1 accuracy: 46.2 % vs 31.6 % | Helped |
| Use MbPA instead of the parametric LSTM | LSTM · Penn Treebank test | perplexity: -5.3 | Helped |
| Combine MbPA with the LSTM baseline and neural cache | LSTM · WikiText-2 test | perplexity: -16.5 | Helped |
| Use MbPA rather than LSTM + cache | LSTM · WikiText-2 test | perplexity: 56.0 vs 51.3 | Helped |
| Use MbPA for test-time dynamic evaluation | LSTM language model · Penn Treebank | perplexity improvement over baseline: gains above 1 perplexity | No clear effect |
| Increase episodic memory size (ablation) | MbPA ImageNet incremental classification · ImageNet novel classes | performance | Helped |
| Retrieve more nearest neighbours for local adaptation (ablation) | MbPA ImageNet incremental classification · ImageNet | performance | Helped |
Closest papers by meaning
- Labeled Memory Networks for Online Model Adaptation (AAAI 2018 · similarity 0.89)
- Dual Memory Networks: A Versatile Adaptation Approach for Vision-Language Models (CVPR 2024 · similarity 0.87)
- Time- Memory- and Parameter-Efficient Visual Adaptation (CVPR 2024 · similarity 0.86)
- Towards flexible perception with visual memory (ICML 2025 · similarity 0.85)
- On the Duality between Gradient Transformations and Adapters (ICML 2025 · similarity 0.85)
- LISA: Layerwise Importance Sampling for Memory-Efficient Large Language Model Fine-Tuning (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.