Research map / Learning theory and algorithms / Sample complexity
Spectral Methods Meet EM: A Provably Optimal Algorithm for Crowdsourcing
Yuchen Zhang, Xi Chen, Dengyong Zhou, Michael I. Jordan
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 spectral initialization instead of majority-voting initialization for Dawid-Skene EM | Dawid-Skene EM after one iteration · Synthetic pi=0.2 | confusion matrix squared error: 5.912 -> 0.972 | Improved |
| use spectral initialization instead of majority-voting initialization | Dawid-Skene EM after 10 iterations · Synthetic | confusion matrix squared error: 0.962 -> 0.962 | No clear effect |
| add one EM iteration after the spectral method (ablation) | Opt-D&S · Synthetic pi=0.2 | confusion matrix squared error: 2.084 -> 0.972 | Improved |
| add EM iterations after spectral method for dense data (ablation) | Opt-D&S · Synthetic pi=1.0 | confusion matrix squared error: 0.164 -> 0.143 | Improved |
| increase sparsity of the data (lower pi) | Opt-D&S · Synthetic | prediction error (%): 0.01 -> 0.84 -> 7.64 (pi 1.0, 0.5, 0.2) | Hurt |
| use Opt-D&S instead of majority voting | majority voting · Synthetic pi=0.2 | prediction error (%): 18.85 -> 7.64 | Improved |
| use Opt-D&S instead of KOS | KOS · TREC | error rate (%): 51.96 -> 29.80 | Improved |
| use Opt-D&S instead of MV-D&S | MV-D&S · Bird | error rate (%): 11.11 -> 10.09 | Improved |
| use Opt-D&S instead of MV-D&S | MV-D&S · Dog | error rate (%): 16.66 -> 16.89 | No clear effect |
| tune the thresholding parameter Delta of spectral initialization (ablation) | Opt-D&S 1st iteration · RTE, Dog, Web | label prediction error: stable after convergence; sensitive after 1st iteration | Improved |
Closest papers by meaning
- Crowdsourcing via Pairwise Co-occurrences: Identifiability and Algorithms (NeurIPS 2019 · similarity 0.92)
- Exact Exponent in Optimal Rates for Crowdsourcing (ICML 2016 · similarity 0.91)
- Analysis of Minimax Error Rate for Crowdsourcing and Its Application to Worker Clustering Model (ICML 2018 · similarity 0.90)
- Optimality of Belief Propagation for Crowdsourced Classification (ICML 2016 · similarity 0.89)
- A Minimax Optimal Algorithm for Crowdsourcing (NeurIPS 2017 · similarity 0.89)
- KFNN: K-Free Nearest Neighbor For Crowdsourcing (NeurIPS 2024 · 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.
