Research map / Continual learning / Continual fine-tuning of LLMs
PACE: Pretrained Audio Continual Learning
Chang Li, Kanglei Zhou, 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 |
|---|---|---|---|
| add improved first-session adaptation (FSA) (ablation) | PACE · TIMIT-2 / TIMIT-3 / VocalSet | accuracy: 75.87 -> 90.95; 75.87 -> 94.05; 61.51 -> 69.08 | Improved |
| add multi-session adaptation (MSA) (ablation) | PACE · TIMIT-2 / TIMIT-3 / VocalSet | accuracy: 85.63 -> 90.95; 89.92 -> 94.05; 62.82 -> 69.08 | Improved |
| add boundary-aware regularization L_reg (ablation) | PACE · TIMIT-2 / TIMIT-3 / VocalSet | accuracy: 89.21 -> 90.95; 93.73 -> 94.05; 66.78 -> 69.08 | Improved |
| add gradient projection (GP) in MSA (ablation) | PACE · TIMIT-2 / TIMIT-3 / VocalSet | accuracy: 88.01 -> 90.95; 89.05 -> 94.05; 58.55 -> 69.08 | Improved |
| use improved FSA instead of naive FSA (ablation) | PACE FSA · SC2 | accuracy: 90.53 -> 91.87 | Improved |
| add low learning rate for FSA head (ablation) | FSA · ESC-50 | accuracy: 92.25 -> 93.75 | Improved |
| add early head freezing to low learning rate (ablation) | FSA · ESC-50 | accuracy: 93.75 -> 94.50 | Improved |
| add layer-aware tuning of later layers (ablation) | FSA · ESC-50 | accuracy: 94.50 -> 95.75 | Improved |
| add naive FSA to RanPAC (ablation) | RanPAC · ESC-50 | accuracy: 92.50 -> 92.25 | Hurt |
| add FSA to RanPAC (ablation) | RanPAC · SC2 | accuracy: 81.22 -> 90.53 | Improved |
| add gradient projection to MSA (ablation) | MSA · VocalSet | forgetting: 57.90 -> 23.69 | Improved |
| replace prompt-based CL with PACE | HiDe-Prompt · TIMIT-2 | accuracy: 47.78 -> 90.95 | Improved |
Closest papers by meaning
- AudioCALM: Continuous Autoregressive Language Modeling for Universal Audio Generation (NeurIPS 2026 · similarity 0.88)
- Continual Audio-Visual Sound Separation (NeurIPS 2024 · similarity 0.88)
- STELLA: Continual Audio-Video Pre-training with SpatioTemporal Localized Alignment (ICML 2024 · similarity 0.88)
- Massive Sound Embedding Benchmark (MSEB) (NeurIPS 2025 · similarity 0.86)
- Weakly-supervised Audio Separation via Bi-modal Semantic Similarity (ICLR 2024 · similarity 0.86)
- Synthio: Augmenting Small-Scale Audio Classification Datasets with Synthetic Data (ICLR 2025 · 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.
