Research map / Label-efficient and robust learning / Recognition and long-tail
ADIOS: Architectures Deep In Output Space
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| two-layer output structure with learned approximate MBC partition (ADIOS) vs flat MLP output | multi-label MLP · Delicious | macro F1: +2.57 (15.12 to 17.69) | Helped |
| two-layer output structure with learned approximate MBC partition (ADIOS) vs flat MLP output | multi-label MLP · MediaMill | macro F1: +6.30 (9.71 to 16.01) | Helped |
| learned MBC partition vs random partition in ADIOS (ablation) | ADIOS · NUS-WIDE | macro F1: +2.73 (11.41 to 14.14) | Helped |
| learned MBC partition vs random partition in ADIOS (ablation) | ADIOS · Delicious | macro F1: +2.22 (15.47 to 17.69) | Helped |
| learned MBC partition vs random partition in ADIOS (ablation) | ADIOS · MediaMill | macro F1: +1.29 (14.72 to 16.01) | Helped |
| learned MBC partition vs random partition in ADIOS (ablation) | ADIOS · BioASQ | macro F1: +0.23 (15.91 to 16.14) | Helped |
| adding G1 labels as input to a second output layer (ADIOS) on NUS-WIDE with P@1 | multi-label MLP · NUS-WIDE | P@1: +4.67 (51.72 to 56.39) | Helped |
| learned MBC partition vs random partition in ADIOS on SUN2012 micro F1 (ablation) | ADIOS · SUN2012 | micro F1: +3.03 (27.29 to 30.32) | Helped |
| random partition ADIOS vs flat MLP on NUS-WIDE (ablation) | ADIOS with random partition · NUS-WIDE | macro F1: -2.59 (14.00 to 11.41) | Hurt |
| learned MBC partition in ADIOS vs flat MLP on Delicious when predicting G2 labels using G1 labels as input (Table 3 predictiveness) | MLP predicting G2 from G1 labels · Delicious | macro F1 on G2: G1 maF1 55.61 vs X 21.54; P@5 X 38.70 vs G1 47.44 | Mixed |
| learned partition on NUS-WIDE: G1 labels vs input features for predicting G2 | MLP predicting G2 · NUS-WIDE | P@5 on G2: X 12.83 vs G1 25.36 P@5; maF1 X 11.93 vs G1 10.64 | Mixed |
Closest papers by meaning
- A no-regret generalization of hierarchical softmax to extreme multi-label classification (NeurIPS 2018 · similarity 0.90)
- Modeling Label Space Interactions in Multi-label Classification using Box Embeddings (ICLR 2022 · similarity 0.88)
- Multi-Label Learning From Single Positive Labels (CVPR 2021 · similarity 0.88)
- Taming the Sigmoid Bottleneck: Provably Argmaxable Sparse Multi-Label Classification (AAAI 2024 · similarity 0.87)
- Self-Paced Unified Representation Learning for Hierarchical Multi-Label Classification (AAAI 2024 · similarity 0.86)
- Harnessing the Power of Beta Scoring in Deep Active Learning for Multi-Label Text Classification (AAAI 2024 · 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.