Research map / Continual learning / Continual fine-tuning of LLMs
Continual Action Quality Assessment via Adaptive Manifold-Aligned Graph Regularization
Kanglei Zhou, Qingyi Pan, Xingxing Zhang, Hubert P. H. Shum, Frederick W. B. Li, Xiaohui Liang, Liyuan Wang
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 |
|---|---|---|---|
| full-parameter fine-tuning instead of a fixed backbone | AQA regression heads (CoRe, HGCN, CoFInAl) · MTL-AQA, AQA-7, FineDiving | SRCC / rMSE: +3.33% SRCC and -31.23% rMSE | Improved |
| full-parameter fine-tuning instead of I3D-Adapter PEFT | AQA regression heads · MTL-AQA, AQA-7, FineDiving | SRCC / rMSE: +2.11% SRCC and -14.25% rMSE | Improved |
| full fine-tuning instead of adapters | AQA regression heads · AQA-7 | SRCC / rMSE: +5.89% SRCC, -44.21% rMSE | Improved |
| full fine-tuning instead of adapters | AQA regression heads · FineDiving | SRCC / rMSE: +0.22% SRCC, -2.85% rMSE | No clear effect |
| full fine-tuning instead of adapters | AQA regression heads · MTL-AQA | SRCC / rMSE: +0.23% SRCC, +4.31% rMSE | Mixed |
| upgrade MAGR to MAGR++ (layer-adaptive FPFT, manifold projector, IIJ-GR) | MAGR · MTL-AQA (default, w/o difficulty label), offline | SRCC rho_avg: 0.8979 -> 0.9205 | Improved |
| upgrade MAGR to MAGR++ | MAGR · FineDiving (w/ dive number), offline | SRCC rho_avg: 0.8580 -> 0.8902 | Improved |
| upgrade MAGR to MAGR++ | MAGR · UNLV-Dive, offline | SRCC rho_avg: 0.7668 -> 0.8165 | Improved |
| upgrade MAGR to MAGR++ | MAGR · MTL-AQA (w/ difficulty), offline | SRCC rho_avg: 0.9237 -> 0.9383 | Improved |
| upgrade MAGR to MAGR++ | MAGR · UNLV-Dive, online | SRCC rho_avg: 0.4202 -> 0.5117 | Improved |
| MAGR++ vs strongest baseline across offline and online CAQA | MAGR++ · four CAQA benchmarks | SRCC rho_avg: 1.6%-6.5% offline and 4.0%-21.8% online; average gains of 3.6% and 12.2% | Improved |
| feature replay methods vs raw-data replay methods | feature replay baselines · MTL-AQA, offline | SRCC rho_avg | Hurt |
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.
- MAGR — MAGR: Manifold-Aligned Graph Regularization for Continual Action Quality Assessment (ECCV 2024) · better on 8 of 8 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
- MAGR: Manifold-Aligned Graph Regularization for Continual Action Quality Assessment (ECCV 2024 · similarity 0.96)
- BriMA: Bridged Modality Adaptation for Multi-Modal Continual Action Quality Assessment (CVPR 2026 · similarity 0.90)
- LucidAction: A Hierarchical and Multi-model Dataset for Comprehensive Action Quality Assessment (NeurIPS 2024 · similarity 0.88)
- FineParser: A Fine-grained Spatio-temporal Action Parser for Human-centric Action Quality Assessment (CVPR 2024 · similarity 0.87)
- ActionAtlas: A VideoQA Benchmark for Domain-specialized Action Recognition (NeurIPS 2024 · similarity 0.85)
- GAIA: Rethinking Action Quality Assessment for AI-Generated Videos (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.
