Research map / Label-efficient and robust learning / Knowledge distillation
Dialogue State Distillation Network with Inter-slot Contrastive Learning for Dialogue State Tracking
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 |
|---|---|---|---|
| use Dialogue State Distillation to learn from previous dialogue states without feeding their predicted values directly (ablation) | DSDN · MultiWOZ 2.0 | Joint GA (%): 53.53 -> 55.30; +1.77 | Helped |
| use Dialogue State Distillation to learn from previous dialogue states without feeding their predicted values directly (ablation) | DSDN · MultiWOZ 2.1 | Joint GA (%): 57.28 -> 59.25; +1.97 | Helped |
| add Inter-slot Contrastive Learning (ablation) | DSDN · MultiWOZ 2.0 | Joint GA (%): 54.25 -> 55.30; +1.05 | Helped |
| add Inter-slot Contrastive Learning (ablation) | DSDN · MultiWOZ 2.1 | Joint GA (%): 58.59 -> 59.25; +0.66 | Helped |
| add Dialogue State Distillation and Inter-slot Contrastive Learning (ablation) | DSDN · MultiWOZ 2.0 | Joint GA (%): 51.93 -> 55.30; +3.37 | Helped |
| add Dialogue State Distillation and Inter-slot Contrastive Learning (ablation) | DSDN · MultiWOZ 2.1 | Joint GA (%): 56.10 -> 59.25; +3.15 | Helped |
| replace independent SOP Cross Entropy Loss with inter-slot Contrastive Learning Loss (ablation) | Baseline · MultiWOZ 2.0 | Joint GA (%): 52.45 -> 53.53; +1.60 vs baseline | Helped |
| replace independent SOP Cross Entropy Loss with inter-slot Contrastive Learning Loss (ablation) | Baseline · MultiWOZ 2.1 | Joint GA (%): 56.76 -> 57.28; +1.18 vs baseline | Helped |
| include same-slot examples from other dialogue turns as contrastive negatives (ablation) | Baseline + Contrastive Learning Loss− · MultiWOZ 2.0 | Joint GA (%): 53.15 -> 53.53; +1.60 vs baseline, compared with +1.22 | Helped |
| include same-slot examples from other dialogue turns as contrastive negatives (ablation) | Baseline + Contrastive Learning Loss− · MultiWOZ 2.1 | Joint GA (%): 57.18 -> 57.28; +1.18 vs baseline, compared with +1.08 | Helped |
| add SOP Cross Entropy Loss to the baseline (ablation) | Baseline · MultiWOZ 2.0 | Joint GA (%): 51.93 -> 52.45; +0.52 | Helped |
| add SOP Cross Entropy Loss to the baseline (ablation) | Baseline · MultiWOZ 2.1 | Joint GA (%): 56.10 -> 56.76; +0.66 | Helped |
Closest papers by meaning
- Schema-Guided Multi-Domain Dialogue State Tracking with Graph Attention Neural Networks (AAAI 2020 · similarity 0.90)
- Learning to Imagine: Distillation-Based Interactive Context Exploitation for Dialogue State Tracking (AAAI 2023 · similarity 0.90)
- TA&AT: Enhancing Task-Oriented Dialog with Turn-Level Auxiliary Tasks and Action-Tree Based Scheduled Sampling (AAAI 2024 · similarity 0.85)
- StateTree: Enhancing Long-term Dialogue Reasoning via Reinforcement Learning (NeurIPS 2026 · similarity 0.84)
- Winnie: Task-Oriented Dialog System with Structure-Aware Contrastive Learning and Enhanced Policy Planning (AAAI 2024 · similarity 0.84)
- Learning to Clarify: Multi-turn Conversations with Action-Based Contrastive Self-Training (ICLR 2025 · 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.