Research map / Probabilistic and scientific modeling / Variational and generative sampling
Data Augmentation for Spoken Language Understanding via Joint Variational Generation
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 |
|---|---|---|---|
| Add JLUVA-generated posterior samples to training data | BiLSTM SLU model · ATIS-small | Slot filling F1: 72.57 -> 74.92 | Helped |
| Add JLUVA-generated posterior samples to training data | BiLSTM SLU model · ATIS-medium | Slot filling F1: 88.28 -> 89.27 | Helped |
| Add JLUVA-generated posterior samples to training data | BiLSTM SLU model · ATIS-full | Slot filling F1: 95.34 -> 95.51 | No clear effect |
| Replace posterior sampling with standard Gaussian prior sampling (ablation) | JLUVA data augmentation for ATIS · ATIS-small | Slot filling F1: 74.92 -> 70.72 | Hurt |
| Replace posterior sampling with standard Gaussian prior sampling (ablation) | JLUVA data augmentation for ATIS · ATIS-medium | Slot filling F1: 89.27 -> 86.90 | Hurt |
| Replace posterior sampling with additive latent perturbation (ablation) | JLUVA data augmentation for ATIS · ATIS-small | Slot filling F1: 74.92 -> 74.14 | No clear effect |
| Augment training data with JLUVA-generated samples | Slot-Gated (Full) · ATIS | Slot F1: 95.3 -> 95.7 | Helped |
| Augment training data with JLUVA-generated samples | Slot-Gated (Full) · Snips | Slot F1: 88.2 -> 89.3 | Helped |
| Augment training data with JLUVA-generated samples | Slot-Gated (Full) · MIT Movie Eng | Slot F1: 82.2 -> 82.9 | Helped |
| Augment training data with JLUVA-generated samples | Slot-Gated (Full) · MIT Movie Trivia | Slot F1: 63.5 -> 65.7 | Helped |
| Augment training data with JLUVA-generated samples | Slot-Gated (Full) · MIT Restaurant | Slot F1: 72.6 -> 73.0 | Helped |
| Increase synthetic-to-real training data ratio (ablation) | JLUVA augmentation with BiLSTM · ATIS-small | Relative performance gain | Mixed |
Closest papers by meaning
- A Variational Framework for Improving Naturalness in Generative Spoken Language Models (ICML 2025 · similarity 0.89)
- Generative Adapter: Contextualizing Language Models in Parameters with A Single Forward Pass (ICLR 2025 · similarity 0.88)
- Improving Variational Encoder-Decoders in Dialogue Generation (AAAI 2018 · similarity 0.88)
- Multimodal Latent Language Modeling with Next-Token Diffusion (ICML 2026 · similarity 0.87)
- Improved Variational Autoencoders for Text Modeling using Dilated Convolutions (ICML 2017 · similarity 0.87)
- Efficient Speech Language Modeling via Energy Distance in Continuous Latent Space (NeurIPS 2025 · 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.