Skip to content

Paper

Two co-authored papers accepted to IEEE GCCE 2026

August 11, 2026

Affectosphere Group

IEEE GCCE 2026

Re-Defining Vanishing Municipalities in Japan: A Multidimensional Clustering and SLM-Based Policy Insight Framework

Toma Okugawa, Keito Inoshita

A reproducible framework (UBRS) that re-defines Japan's 'vanishing municipalities' beyond a single demographic axis — clustering 1,721 municipalities across 125 indicators (UMAP + HDBSCAN + SHAP) and injecting each cluster profile as RAG context into a locally deployable small language model to generate municipality-tailored policy insights.

IEEE GCCE 2026

Behavioral Fidelity and Philosophy-Grounded Design for Self-Reflective Conversational AI

Takumi Matsuo, Keito Inoshita

Proposes Behavioral Fidelity (BF), a framework that quantifies whether an LLM dialogue agent actually follows its designer-specified behavioral rules — via LLM-as-judge strategy classification, KL-divergence analysis, and sequential constraint checks — and applies it to Mirra, a philosophy-of-language-grounded self-reflection app.