Research map / Reinforcement learning / RL from feedback
AvgOut: A Simple Output-Probability Measure to Eliminate Dull Responses
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 |
|---|---|---|---|
| train with MinAvgOut regularization of AvgOut | LSTM dialogue model · Ubuntu Dialogue | Activity F1: 5.94 -> 6.83 | Helped |
| train with MinAvgOut regularization of AvgOut | LSTM dialogue model · Ubuntu Dialogue | iAUC-avg: 0.6893 -> 0.7305 | Helped |
| condition generation on a diversity label scaled to target diversity | LSTM dialogue model · Ubuntu Dialogue | Activity F1: 5.94 -> 7.55 | Helped |
| condition generation on a diversity label scaled to target diversity | LSTM dialogue model · Ubuntu Dialogue | Distinct-1: 0.0076 -> 0.0056 | Hurt |
| use AvgOut diversity score as sampled-response reinforcement-learning reward | LSTM dialogue model · Ubuntu Dialogue | Activity F1: 5.94 -> 5.80 | No clear effect |
| combine MinAvgOut and RL losses | LSTM dialogue model · Ubuntu Dialogue | Activity F1: 6.83 -> 8.05 | Helped |
| combine MinAvgOut and RL losses | LSTM dialogue model · Ubuntu Dialogue | iAUC-avg: 0.7305 -> 0.7200 | Hurt |
| combine MinAvgOut and RL losses | LSTM dialogue model · Ubuntu Dialogue | Avg. human score: 3.48 -> 3.45 | No clear effect |
| encourage unigram diversity through AvgOut regularization | MinAvgOut · Ubuntu Dialogue | iAUC-2: 0.725 -> 0.754 | Helped |
| use response diversity as the RL reward | RL · Ubuntu Dialogue | Content Richness: 2.88 | Hurt |
Closest papers by meaning
- DialogBERT: Discourse-Aware Response Generation via Learning to Recover and Rank Utterances (AAAI 2021 · similarity 0.88)
- SD-Eval: A Benchmark Dataset for Spoken Dialogue Understanding Beyond Words (NeurIPS 2024 · similarity 0.87)
- SimpleStrat: Diversifying Language Model Generation with Stratification (NeurIPS 2025 · similarity 0.87)
- WildBench: Benchmarking LLMs with Challenging Tasks from Real Users in the Wild (ICLR 2025 · similarity 0.86)
- XTC: Head-Aware Sampling by Excluding Top Choices (NeurIPS 2026 · similarity 0.86)
- A Comprehensive Analysis of the Effectiveness of Large Language Models as Automatic Dialogue Evaluators (AAAI 2024 · 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.