Research map / Architectures and efficiency / Attention and transformers
An Online Sequence-to-Sequence Model Using Partial Conditioning
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 |
|---|---|---|---|
| maintain transducer recurrent state across blocks (ablation) | Neural Transducer · TIMIT, W=15 | PER: 34.3 -> 20.6 | Helped |
| use Kaldi GMM-HMM alignments instead of model-generated alignments (ablation) | Neural Transducer · TIMIT test | PER: 20.8 -> 19.8 | Helped |
| use 3-layer bidirectional encoder instead of unidirectional (ablation) | Neural Transducer · TIMIT test | PER: 19.8 -> 18.7 | Helped |
| compare to best previous seq2seq model | Neural Transducer · TIMIT test | PER: 18.7 -> 17.6 | Hurt |
| choose window size W=8 for no-attention transducer (ablation) | Neural Transducer without attention · TIMIT | PER | Helped |
| add attention (DOT, MLP, LSTM) to transducer (ablation) | Neural Transducer · TIMIT | PER | No clear effect |
| LSTM attention with W=25 (ablation) | Neural Transducer · TIMIT | PER | Helped |
| increase encoder depth from 2 to 3 layers (ablation) | Neural Transducer · TIMIT | PER: 18.8 -> 18.5 | Helped |
| four-layer transducer (ablation) | Neural Transducer · TIMIT | PER | Mixed |
Closest papers by meaning
- Sequence to Multi-Sequence Learning via Conditional Chain Mapping for Mixture Signals (NeurIPS 2020 · similarity 0.86)
- Trainable Decoding of Sets of Sequences for Neural Sequence Models (ICML 2019 · similarity 0.86)
- Dynamic Evaluation of Neural Sequence Models (ICML 2018 · similarity 0.85)
- Predictive Attractor Models (NeurIPS 2024 · similarity 0.84)
- Efficient Sequence Transduction by Jointly Predicting Tokens and Durations (ICML 2023 · similarity 0.84)
- Proteus: Incremental Memory Activation for Long-Context Sequence Modeling (NeurIPS 2026 · similarity 0.84)
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.