Research.
Evolutionary computation, stochastic optimization and evaluation of LLM-based dialogue agents. Selected publications and presentations.
Research & publications
Research topics
- Evolutionary computation & stochastic optimization
- Searching for solutions that meet multiple evaluation criteria.
- Evaluating LLMs & dialogue agents
- Separating an agent’s specified persona from its conversation history to evaluate changes in its responses.
Selected publications & presentations
HAI 2026 · Peer-reviewed conference paper · Published 15 November 2026
Persona-Isolated Evaluation of LLM-Based Dialogue Agents: Measuring History-Conditioned Response Variation
Satoshi Koda, Hiroshi Someya, Kagayaki Kuroda
Proceedings of the 14th International Conference on Human-Agent Interaction, pp. 20–28. ACM.
How does a dialogue agent answer the same question after different conversations? We propose an evaluation method that keeps conversation histories separate for each persona and compares how large language models (LLMs) respond to the same prompts.
2022 · IEEJ Systems Technical Meeting / ST-22-037
方向ベクトル集合により表現される解候補の評価関数定式化
Hiroshi Someya, Satoshi Koda
A study of solution-evaluation functions based on directional statistics.
Publisher record2019 · IEEJ Electronics, Information and Systems Conference / TC6-5
遺伝的アルゴリズムによる芳香療法のための精油調合パターン最適化における共起解析
Satoshi Koda, Hiroshi Someya
Analysis of co-occurrence relationships among essential oils in blends produced by multi-objective optimization.
Publisher record