Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Sequence Classification using Sequential Output Statistics
Yu Liu, Jianshu Chen, Li Deng
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 SPDG to optimize Empirical-ODM instead of SGD with batch size 10k | Linear sequence classifier · OCR test | error rate: 56.48% -> 9.59% | Improved |
| Use SPDG to optimize Empirical-ODM instead of SGD with batch size 10k | Linear sequence classifier · Spell-Corr test | error rate: 45.24% -> 1.94% | Improved |
| Replace the method from [cite] with Empirical-ODM optimized by SPDG | Linear sequence classifier · OCR test | error rate: 83.37% -> 9.59% | Improved |
| Replace the method from [cite] with Empirical-ODM optimized by SPDG | Linear sequence classifier · Spell-Corr test | error rate: 82.91% -> 1.94% | Improved |
| Add sequential output statistics by using 2-gram or 3-gram rather than 1-gram (ablation) | Empirical-ODM · OCR test, all four LM sources | error rate: 71.83% -> 10.93%; 71.83% -> 10.17%; 72.14% -> 12.55%; 72.14% -> 12.89%; 71.51% -> 10.56%; 71.51% -> 10.29 %; 71.25% -> 1 | Improved |
| Use 3-gram instead of 2-gram LM (ablation) | Empirical-ODM · OCR test, all four LM sources | error rate: 10.93% -> 10.17%; 12.55% -> 12.89%; 10.56% -> 10.29 %; 10.33% -> 9.21% | Mixed |
| Use Empirical-ODM's coverage-seeking cost instead of the earlier mode-seeking cost | Unsupervised sequence classifier · OCR test | error rate: 83.37% -> 9.59% | Improved |
Closest papers by meaning
- A Sequential Set Generation Method for Predicting Set-Valued Outputs (AAAI 2019 · similarity 0.87)
- Incremental Sequence Classification with Temporal Consistency (NeurIPS 2025 · similarity 0.85)
- Parallelizing non-linear sequential models over the sequence length (ICLR 2024 · similarity 0.84)
- On the Provable Advantage of Unsupervised Pretraining (ICLR 2024 · similarity 0.84)
- Progressive Inference: Explaining Decoder-Only Sequence Classification Models Using Intermediate Predictions (ICML 2024 · similarity 0.84)
- Realistic Unsupervised CLIP Fine-tuning with Universal Entropy Optimization (ICML 2024 · 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.
