Research map / Label-efficient and robust learning / Recognition and long-tail
Learnability with Indirect Supervision Signals
Kaifu Wang, Qiang Ning, Dan Roth
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 indirect supervision with a transition class whose induced label distributions are separated | Multiclass classification from indirect supervision · Realizable multiclass classification | learnability | Improved |
| Restrict the transition class using prior information that separates label-induced annotation distributions | Multiclass classification from indirect supervision · Realizable multiclass classification | identifiability: η ≥ γ > 0 | Improved |
| Use a transition class with no positive separation | Multiclass classification from indirect supervision · Realizable multiclass classification | identifiability: lim_k η^(k)=0 | Hurt |
| Use concentration sets that distinguish the annotation distributions for each label | Multiclass classification from indirect supervision · Concentration condition | separation degree: γ ≥ 2γ_C^2 | Improved |
| Use the transition-independent loss 1{o not in S_h(x)} under the concentration condition | ERM for indirect supervision · Concentration condition | identifiability: η ≥ γ_C | Improved |
| Combine two annotation sources without constraints on their joint transition | Joint supervision · Pairwise label separation | separation degree: γ_i→j ≤ λγ_i→j1 + (1−λ)γ_i→j2 | Mixed |
| Distinguish the identities of annotation sources in joint supervision | Joint supervision · Pairwise label separation | separation degree: γ_i→j = λγ_i→j1 + (1−λ)γ_i→j2 | Improved |
| Mix annotation sources without retaining source identity | Joint supervision · Binary classification with two annotators | separation: λ=1/2 | Hurt |
| Impose prior knowledge that one annotator has better quality than the other | Joint supervision with two noisy annotators · Binary classification | learnability: η_1−η_2≤γ<0 | Improved |
Closest papers by meaning
- On Learning Latent Models with Multi-Instance Weak Supervision (NeurIPS 2023 · similarity 0.85)
- Learning with Partial-Label and Unlabeled Data: A Uniform Treatment for Supervision Redundancy and Insufficiency (ICML 2024 · similarity 0.85)
- Creating Training Sets via Weak Indirect Supervision (ICLR 2022 · similarity 0.85)
- Labels, Information, and Computation: Efficient Learning Using Sufficient Labels (JMLR 2023 · similarity 0.85)
- Imprecise Label Learning: A Unified Framework for Learning with Various Imprecise Label Configurations (NeurIPS 2024 · similarity 0.84)
- Instance-wise Supervision-level Optimization in Active Learning (CVPR 2025 · similarity 0.83)
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.
