Research map / Label-efficient and robust learning / Recognition and long-tail
Deep Streaming Label Learning
Zhen Wang, Liu Liu, Dacheng Tao
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 streaming feature distillation (KD) to the plain deep network (ablation) | DSLL new-label classifier · MirFlickr, 9 new labels | average precision: 0.3824 to 0.4018 | Improved |
| replace cross-entropy with label correlation-aware loss (LC) in streaming label mapping (ablation) | streaming label mapping · MirFlickr, 9 new labels | average precision: 0.4113 to 0.4182 (slight) | Improved |
| add transformation layer St after streaming label mapping (ablation) | LC label mapping · MirFlickr, 9 new labels | average precision: 0.4182 to 0.4287 | Improved |
| use both hard and soft past labels as mapping input vs hard only (ablation) | streaming label mapping · MirFlickr, 9 new labels | average precision: Lhard_soft 0.4592 vs Lhard 0.4159 but Lsoft 0.4427 is higher than Lhard | Mixed |
| full DSLL (KD + LC-St + hard/soft mapping) vs plain baseline (ablation) | DSLL · MirFlickr, 9 new labels | average precision: 0.3824 to 0.4925 | Improved |
| full DSLL vs plain baseline on coverage (ablation) | DSLL · MirFlickr, 9 new labels | coverage: 0.2415 to 0.2153 (lower is better) | Improved |
| full DSLL vs strongest baseline (C2AE) | DSLL · Wiki10 and MirFlickr, micro-AUC and ranking loss | micro-AUC / ranking loss: Wiki10 1k micro-AUC 0.8406 vs 0.8274; ranking loss 0.0421 vs 0.0483 | Improved |
| full DSLL vs SLL/SML and C2AE on EURlex and Delicious | DSLL · EURlex 1000 new labels micro-AUC | micro-AUC: DSLL 0.9106 vs SLL 0.9115 | Mixed |
Closest papers by meaning
- Collaborative Learning of Label Semantics and Deep Label-Specific Features for Multi-Label Classification (TPAMI 2021 · similarity 0.88)
- ADIOS: Architectures Deep In Output Space (ICML 2016 · similarity 0.88)
- The Emerging Trends of Multi-Label Learning (TPAMI 2021 · similarity 0.88)
- Deep Determinantal Point Process for Large-Scale Multi-Label Classification (ICCV 2017 · similarity 0.87)
- CNN-RNN: A Unified Framework for Multi-Label Image Classification (CVPR 2016 · similarity 0.87)
- Two-Stage Label Embedding via Neural Factorization Machine for Multi-Label Classification (AAAI 2019 · similarity 0.87)
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.
