Research map / Continual learning / Continual fine-tuning of LLMs
DSCA: Dynamic Subspace Concept Alignment for Lifelong VLM Editing
Gyanendra Das, Sai Jena
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 orthogonality constraint in subspaces (ablation) | DSCA · Ablation | Edit Success: 95.8 -> 98.0 | Improved |
| add orthogonality constraint in subspaces (ablation) | DSCA · Ablation | Locality drop: 2.8 -> 0.5 | Improved |
| add orthogonality constraint in subspaces (ablation) | DSCA · Ablation | Generalization: 93.4 -> 97.3 | Improved |
| add gate sparsity loss (ablation) | DSCA · Ablation | Edit Success: 96.1 -> 98.0 | Improved |
| add gate sparsity loss (ablation) | DSCA · Ablation | Locality drop: 2.1 -> 0.5 | Improved |
| use two-stage hierarchical routing instead of single stage (ablation) | DSCA · Ablation | Locality drop: 1.9 -> 0.5 | Improved |
| add basis-residual intervention (ablation) | DSCA · Ablation | Locality drop: 1.5 -> 0.5 | Improved |
| use full number of subspaces K instead of K/2 (ablation) | DSCA · Ablation | Edit Success: 97.0 -> 98.0 | Improved |
| use full rank r instead of r/2 (ablation) | DSCA · Ablation | Locality drop: 1.2 -> 0.5 | Improved |
| residualized incremental PCA vs no orthogonalization (ablation) | DSCA · 1000 sequential edits | subspace overlap: 10^-1 -> 3x10^-3 | Improved |
| apply DSCA instead of LiveEdit in lifelong editing | DSCA · E-VQA lifelong (1000 edits) | Rel.: 92.93 -> 96.85 | Improved |
| apply DSCA instead of PAM on CoIN | DSCA · CoIN | BWT: -19.45 -> -9.37 | Improved |
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.
- MagMax — MagMax: Leveraging Model Merging for Seamless Continual Learning (ECCV 2024) · better on 2 of 2 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
- Lifelong Knowledge Editing for Vision Language Models with Low-Rank Mixture-of-Experts (CVPR 2025 · similarity 0.90)
- MemEIC: A Step Toward Continual and Compositional Knowledge Editing (NeurIPS 2025 · similarity 0.89)
- VLKEB: A Large Vision-Language Model Knowledge Editing Benchmark (NeurIPS 2024 · similarity 0.88)
- WISE: Rethinking the Knowledge Memory for Lifelong Model Editing of Large Language Models (NeurIPS 2024 · similarity 0.87)
- Attribution Analysis Meets Model Editing: Advancing Knowledge Correction in Vision Language Models with VisEdit (AAAI 2025 · similarity 0.87)
- Gated Relational Alignment via Confidence-based Distillation for Efficient VLMs (ICML 2026 · similarity 0.86)
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.
