Research map / Continual learning / Online and replay-based
Online Noisy Continual Relation Learning
Guozheng Li, Peng Wang, Qiqing Luo, Yanhe Liu, Wenjun Ke
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 |
|---|---|---|---|
| Use S6 instead of the strongest reported baseline | FewRel, TACRED, and NYT-H · FewRel at 20%, 40%, 60%, and 80% noise; TACRED at 20%, 40%, 60%, and 80% noise; NYT-H | overall accuracy (FewRel and TACRED) or macro-F1 (NYT-H): FewRel: 64.6 -> 66.1; 55.3 -> 66.0; 36.9 -> 58.8; 17.8 -> 50.1. TACRED: 78.4 -> 61.2; 47.7 -> 43.9; 18.2 -> 24.3; 8.6 -> | Improved |
| Add noisy replay-buffer samples to clean samples during semi-supervised offline finetuning (ablation) | S6 · FewRel, 80% noise | overall accuracy (%): 49.6 -> 50.1 | Improved |
| Add noisy replay-buffer samples to clean samples during semi-supervised offline finetuning (ablation) | S6 · TACRED, 80% noise | overall accuracy (%): 10.8 -> 12.2 | Improved |
| Add noisy replay-buffer samples to clean samples during semi-supervised offline finetuning (ablation) | S6 · NYT-H | macro-F1 (%): 20.4 -> 29.1 | Improved |
| Add self-supervised contrastive online learning instead of supervised contrastive learning (ablation) | S6 · FewRel and TACRED across noise rates | accuracy | Improved |
| Replace the sample separator's loss-based selection rules with random sampling (ablation) | S6 replay-buffer selection · FewRel, first-task accuracy over subsequent tasks | accuracy | Hurt |
| Add the self-supervised online learner to GDumb's greedy sampler (ablation) | GDumb · FewRel and TACRED across noise rates | accuracy | Improved |
| Use noisy replay samples in finetuning rather than only clean replay samples (ablation) | S6 · FewRel, 80% noise | overall accuracy (%): 49.1 -> 50.1 | Improved |
| Use noisy replay samples in finetuning rather than only clean replay samples (ablation) | S6 · TACRED, 80% noise | overall accuracy (%): 10.5 -> 12.2 | Improved |
| Use noisy replay samples in finetuning rather than only clean replay samples (ablation) | S6 · NYT-H | macro-F1 (%): 18.2 -> 29.1 | Improved |
| Apply loss-based sample purification on an extremely imbalanced dataset (ablation) | S6 · NYT-H clean replay buffer | clean-sample proportion: 34.5 | 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.
- GDumb — GDumb: A Simple Approach that Questions Our Progress in Continual Learning (ECCV 2020) · 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
- Continual Relation Extraction via Sequential Multi-Task Learning (AAAI 2024 · similarity 0.87)
- Improving Plasticity in Online Continual Learning via Collaborative Learning (CVPR 2024 · similarity 0.85)
- Dynamic Sub-graph Distillation for Robust Semi-supervised Continual Learning (AAAI 2024 · similarity 0.85)
- Few-Shot, No Problem: Descriptive Continual Relation Extraction (AAAI 2025 · similarity 0.85)
- Continual Learning on a Diet: Learning from Sparsely Labeled Streams Under Constrained Computation (ICLR 2024 · similarity 0.85)
- Non-exemplar Online Class-Incremental Continual Learning via Dual-Prototype Self-Augment and Refinement (AAAI 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.
