Research map / Continual learning / Continual fine-tuning of LLMs
Learning Adaptive and Expandable Mixture Model for Continual Learning
Fei Ye, YongCheng Zhong, Qihe Liu, Adrian G. Bors, JingLing Sun, Jinyu Guo, ShiJie Zhou
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 prediction calibration process (PCP) (ablation) | AEMM (mse) · MTIL 3 task orders (TIN/CD/RESISC45), mean over orders | Average accuracy: 89.65 -> 90.15 | Improved |
| add representation calibration process (RCP) (ablation) | AEMM (mse) · MTIL 3 task orders (TIN/CD/RESISC45), mean over orders | Average accuracy: 89.77 -> 90.15 | Improved |
| add PCP (ablation) | AEMM (mse) · MTIL 3 task orders (TIN/CD/RESISC45), mean over orders | Last accuracy: 89.56 -> 90.03 | Improved |
| add RCP (ablation) | AEMM (mse) · MTIL 3 task orders (TIN/CD/RESISC45), mean over orders | Last accuracy: 89.56 -> 90.03 | Improved |
| add PCP (ablation) | AEMM (mse) · MTIL TIN-CD-RESISC45 order | Average accuracy: 86.35 -> 86.91 | Improved |
| add RCP (ablation) | AEMM (mse) · MTIL TIN-CD-RESISC45 order | Average accuracy: 86.19 -> 86.91 | Improved |
| add PCP (ablation) | AEMM (mse) · MTIL CD-RESISC45-TIN order | Average accuracy: 93.74 -> 94.29 | Improved |
| add RCP (ablation) | AEMM (mse) · MTIL CD-RESISC45-TIN order | Average accuracy: 94.18 -> 94.29 | Improved |
| use MSE instead of MMD for representation calibration | AEMM · MTIL (7 datasets), Total Avg | Average accuracy: 87.60 -> 87.77 | No clear effect |
| use AEMM dual-representation expandable model instead of strongest baseline CLS-ER | AEMM (mse) · MTIL (7 datasets), Total Avg | Average accuracy: 81.00 -> 87.77 | Improved |
| increase PCP coefficient (ablation) | AEMM (RCP=0) · MTIL (7 datasets) | Average accuracy: peaking at 87.58% | Improved |
| increase RCP coefficient beyond 0.3 (ablation) | AEMM (PCP=0) · MTIL (7 datasets) | Average accuracy | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Ranpac — RanPAC: Random Projections and Pre-trained Models for Continual Learning (NeurIPS 2023) · better on 3 of 3 shared results
- DualPrompt — DualPrompt: Complementary Prompting for Rehearsal-Free Continual Learning (ECCV 2022) · better on 2 of 2 shared results
- SLCA — SLCA++: Unleash the Power of Sequential Fine-Tuning for Continual Learning with Pre-training (IJCV 2026) · better on 1 of 1 shared result
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
- Self-Expansion of Pre-trained Models with Mixture of Adapters for Continual Learning (CVPR 2025 · similarity 0.90)
- BECAME: Bayesian Continual Learning with Adaptive Model Merging (ICML 2025 · similarity 0.87)
- Improving Continual Learning Performance and Efficiency with Auxiliary Classifiers (ICML 2025 · similarity 0.87)
- Online Task-Free Continual Learning via Dynamic Expansionable Memory Distribution (CVPR 2025 · similarity 0.87)
- Learning to Continually Learn with the Bayesian Principle (ICML 2024 · similarity 0.87)
- SAFE: Slow and Fast Parameter-Efficient Tuning for Continual Learning with Pre-Trained Models (NeurIPS 2024 · similarity 0.87)
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.
