Blog
AI systems · Agent infrastructure · Network Science × RecSys
-
The Shell Is a Protocol NowWhy an agent-native CLI is a stable execution and governance boundary—and why precise, executable requirements are what make autonomous exploration safe.
-
Swarm-Coder 搭建日志：从零构建低成本多SLM代码生成集群用 Network Science 思维打造一个「蜂群」架构。Phase 1 完成: LangGraph workflow, popularity bias penalty 路由算法, Gini 系数指标。低成本 vLLM + Ollama 方案。
-
Anatomy of a Recommendation GraphTaking MovieLens-1M apart as a network: heavy-tailed degree distributions, Lorenz curves for popularity inequality, how SAR's similarity metrics are really network normalization schemes, and why the Recommenders project needs popularity bias metrics.
-
Measuring What Matters: Fairness and Popularity Bias in Recommender SystemsRecommender systems amplify popularity through the same preferential attachment that creates scale-free networks. Exploring 10 fairness and bias metrics — Gini index, calibration error, exposure fairness — with math, implementation, and connections to network science.
-
Why Networks Matter for RecommendationsEvery recommendation system is secretly a network problem. Exploring user-item bipartite graphs, power-law degree distributions, community structure, link prediction, and the spectral theory behind GNN-based recommenders.