Research map / Continual learning / Class-incremental
Time Series Class-Incremental Learning via Confidence-guided Mask Distillation and Prototype-guided Contrastive Learning
Yu Liu, Haoqin Yang, Jinping Sui, Hui Wang, Haipeng Li, Weimin Wang, Qi Jia
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 confidence-guided mask distillation (CMD) to DT2W baseline (ablation) | DT2W baseline · UCI-HAR | average accuracy / forgetting: A_T 51.50 -> 55.42; F_T 62.00 -> 47.86 | Improved |
| add prototype-guided contrastive learning (PCL) alone to DT2W baseline (ablation) | DT2W baseline · UCI-HAR | average accuracy / forgetting: A_T 51.50 -> 52.04; F_T 62.00 -> 63.79 | Mixed |
| combine CMD and PCL (ablation) | DT2W baseline · UCI-HAR | average accuracy / forgetting: A_T 51.50 -> 59.91; F_T 62.00 -> 44.55 | Improved |
| reduce lambda1 (weight of masked feature distillation) to 0.001 (ablation) | CMD+PCL · UCI-HAR | average accuracy / forgetting: A_T 59.8 -> 33.6; F_T 47.8 -> 93.1 | Hurt |
| increase lambda1 to 1.0 (ablation) | CMD+PCL · UCI-HAR | average accuracy: 59.8 -> 44.9 | Hurt |
| increase lambda2 (weight of PCL loss) to 1.0 (ablation) | CMD+PCL · UCI-HAR | average accuracy / forgetting: A_T 59.8 -> 48.5; F_T 47.8 -> 69.9 | Hurt |
| decrease lambda2 to 0.01 (ablation) | CMD+PCL · UCI-HAR | average accuracy / forgetting: A_T 59.8 -> 58.3; F_T 47.8 -> 56.2 | Hurt |
| apply CMD+PCL versus best replay-free baseline DT2W | 1D-CNN replay-free TSCIL · DSA | average accuracy / forgetting: A_T 24.5 -> 29.6; F_T 87.2 -> 60.6 | Improved |
| apply CMD+PCL versus DT2W on WISDM | 1D-CNN replay-free TSCIL · WISDM | average accuracy / forgetting: A_T 16.4 -> 19.0; F_T 62.1 -> 57.3 | Improved |
| add PCL loss, visualised with t-SNE (ablation) | CMD · UCI-HAR | feature separability (qualitative) | 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.
- MIND — MIND: Multi-Task Incremental Network Distillation (AAAI 2024) · 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
- Language Pre-training Guided Masking Representation Learning for Time Series Classification (AAAI 2025 · similarity 0.86)
- Frequency-Masked Embedding Inference: A Non-Contrastive Approach for Time Series Representation Learning (AAAI 2025 · similarity 0.85)
- eTag: Class-Incremental Learning via Embedding Distillation and Task-Oriented Generation (AAAI 2024 · similarity 0.85)
- Prediction Error-based Classification for Class-Incremental Learning (ICLR 2024 · similarity 0.85)
- Abstain Mask Retain Core: Time Series Prediction by Adaptive Masking Loss with Representation Consistency (NeurIPS 2025 · similarity 0.85)
- Mixture of Noise for Pre-Trained Model-Based Class-Incremental Learning (NeurIPS 2025 · similarity 0.85)
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.
