Research map / Continual learning / Continual fine-tuning of LLMs
D-VLAM: Differential Vision and Language Mixing for Rehearsal Free Continual Learning
Muhammad Anwar Ma'sum, Mohsen Guizani, Waseem Ullah
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 differential prefix tuning (lambda_t + DPT) (ablation) | D-VLAM · ImageNet-R | FAA: 85.5 -> 86.5 | Improved |
| add language-modality prompt generator Gt(E) (ablation) | D-VLAM · ImageNet-R | FAA: 82.2 -> 85.5 | Improved |
| add class-level keys and loss (ablation) | D-VLAM · ImageNet-R | FAA: 76.8 -> 82.2 | Improved |
| add task-level key and task-wise generators (ablation) | D-VLAM · ImageNet-R | FAA: 71.9 -> 76.8 | Improved |
| add vision prompt generator G(x) (ablation) | fine-tuning · ImageNet-R | FAA: 59.5 -> 71.9 | Improved |
| add inter-task loss (ablation) | fine-tuning · ImageNet-R | FFM: FAA 58.5 -> 59.5; FFM 5.9 -> 7.1 | Mixed |
| complementary-LV or aggregate instead of complementary-VL mixing (ablation) | D-VLAM · ImageNet-R | FAA: 86.5 vs 86.1 / 86.4 | No clear effect |
| add DPT to each mixing configuration (ablation) | D-VLAM · ImageNet-R | FAA: 85.4 -> 86.1 (Comp-LV) | Improved |
| use D-VLAM instead of LEAPGen | language-guided prompt CL · ImageNet-R 10-task | FAA: 83.18 -> 86.54 | Improved |
| D-VLAM vs LEAPGen training time | D-VLAM · ImageNet-R 10-task | training time (h): 0.655 -> 0.725 | Hurt |
| use class-name text and lightweight generators instead of larger generators | D-VLAM · ImageNet-R 10-task | trainable params: 6.35 M -> 0.16 M (vs LEAPGen) | Improved |
Closest papers by meaning
- Synthetic Data is an Elegant GIFT for Continual Vision-Language Models (CVPR 2025 · similarity 0.90)
- Dynamic Cross-Modal Prompt Generation for Multimodal Continual Instruction Tuning (NeurIPS 2026 · similarity 0.89)
- Enhancing Visual Continual Learning with Language-Guided Supervision (CVPR 2024 · similarity 0.89)
- Convolutional Prompting meets Language Models for Continual Learning (CVPR 2024 · similarity 0.89)
- Stabilizing Zero-Shot Prediction: A Novel Antidote to Forgetting in Continual Vision-Language Tasks (NeurIPS 2024 · similarity 0.89)
- Overcoming Catastrophic Forgetting in Visual Continual Learning with Reinforcement Fine-Tuning (NeurIPS 2026 · 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.
