Research map / Reinforcement learning / Imitation learning
Dialogue Generation: From Imitation Learning to Inverse Reinforcement Learning
Ziming Li, Julia Kiseleva, Maarten de Rijke
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 a state-action reward model with importance sampling instead of the DG-AIL discriminator reward (ablation) | DG-AIRL dialogue generation · MovieTriples human pairwise vs DG-AIL | Win rate: 0.36 win, 0.37 tie, 0.27 loss | Mixed |
| Add causal entropy regularization to adversarial training (ablation) | SeqGan · MovieTriples human pointwise | Average score: 0.39 -> 0.53 | Improved |
| Add causal entropy regularization to adversarial training (ablation) | SeqGan · MovieTriples human pairwise vs DG-AIRL | Win rate: 0.28 loss -> 0.27 loss | Improved |
| Replace the adversarial imitation reward with a learned state-action reward and importance sampling (ablation) | DG-AIL · MovieTriples test | Average embedding: 0.553 ± 0.003 -> 0.589* ± 0.003 | Improved |
| Replace the adversarial imitation reward with a learned state-action reward and importance sampling (ablation) | DG-AIL · MovieTriples human pointwise | Average score: 0.53 -> 0.54 | Improved |
| Replace the adversarial imitation reward with a learned state-action reward and importance sampling (ablation) | DG-AIL · MovieTriples human pairwise | Win rate: 0.36 win, 0.37 tie, 0.27 loss | Mixed |
| Use DG-AIRL instead of Seq2Seq | Dialogue generation · MovieTriples human pairwise | Win rate: 0.44 win, 0.29 tie, 0.27 loss | Improved |
| Use DG-AIRL instead of VHRED | Dialogue generation · MovieTriples human pairwise | Win rate: 0.46 win, 0.32 tie, 0.22 loss | Improved |
| Use DG-AIRL instead of SeqGan | Dialogue generation · MovieTriples human pairwise | Win rate: 0.47 win, 0.25 tie, 0.28 loss | Improved |
Closest papers by meaning
- Are You Talking to Me? Reasoned Visual Dialog Generation Through Adversarial Learning (CVPR 2018 · similarity 0.88)
- Inverse Reinforcement Learning with Natural Language Goals (AAAI 2021 · similarity 0.87)
- GPT-Critic: Offline Reinforcement Learning for End-to-End Task-Oriented Dialogue Systems (ICLR 2022 · similarity 0.87)
- Re-Evaluating ADEM: A Deeper Look at Scoring Dialogue Responses (AAAI 2019 · similarity 0.86)
- Posterior-GAN: Towards Informative and Coherent Response Generation with Posterior Generative Adversarial Network (AAAI 2020 · similarity 0.86)
- Imitating Language via Scalable Inverse Reinforcement Learning (NeurIPS 2024 · similarity 0.85)
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.
