Research map / Adversarial robustness and security / Certified and empirical robustness
Generating Informative and Diverse Conversational Responses via Adversarial Information Maximization
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 adversarial training to seq2seq (cGAN) (ablation) | CNN-LSTM response generation · Reddit test | Dist-2: 0.153 -> 0.199 | Helped |
| Add adversarial training to seq2seq (cGAN) (ablation) | CNN-LSTM response generation · Twitter test | BLEU: 0.64 -> 0.62 | Hurt |
| Add the variational mutual-information objective to cGAN (AIM) (ablation) | cGAN · Reddit test | BLEU: 1.83 -> 2.04 | Helped |
| Add the variational mutual-information objective to cGAN (AIM) (ablation) | cGAN · Reddit test | Dist-2: 0.199 -> 0.205 | Helped |
| Add the variational mutual-information objective to cGAN (AIM) (ablation) | cGAN · Twitter test | BLEU: 0.62 -> 0.85 | Helped |
| Add the variational mutual-information objective to cGAN (AIM) (ablation) | cGAN · Twitter test | Dist-2: 0.102 -> 0.092 | Hurt |
| Add the dual forward-backward adversarial objective (DAIM) to AIM (ablation) | AIM · Reddit test | Dist-2: 0.205 -> 0.220 | Helped |
| Add the dual forward-backward adversarial objective (DAIM) to AIM (ablation) | AIM · Reddit test | BLEU: 2.04 -> 1.93 | Hurt |
| Add the dual forward-backward adversarial objective (DAIM) to AIM (ablation) | AIM · Twitter test | Dist-2: 0.092 -> 0.137 | Helped |
| Add the dual forward-backward adversarial objective (DAIM) to AIM (ablation) | AIM · Twitter test | BLEU: 0.85 -> 0.81 | Hurt |
| Use DAIM rather than MMI | Conversational response generation · Reddit human evaluation | informativeness preference: 0.484 vs 0.516 | Helped |
| Use DAIM rather than MMI | Conversational response generation · Reddit human evaluation | relevance preference: 0.503 vs 0.497 | No clear effect |
Closest papers by meaning
- SimpleStrat: Diversifying Language Model Generation with Stratification (NeurIPS 2025 · similarity 0.86)
- Measuring Non-Adversarial Reproduction of Training Data in Large Language Models (ICLR 2025 · similarity 0.85)
- REINFORCE Adversarial Attacks on Large Language Models: An Adaptive, Distributional, and Semantic Objective (ICML 2025 · similarity 0.84)
- LIDAO: Towards Limited Interventions for Debiasing (Large) Language Models (ICML 2024 · similarity 0.84)
- Rainbow Teaming: Open-Ended Generation of Diverse Adversarial Prompts (NeurIPS 2024 · similarity 0.83)
- A Multi-Agent Conversational Bandit Approach to Online Evaluation and Selection of User-Aligned LLM Responses (AAAI 2026 · similarity 0.83)
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.