PyImpetus is a Markov Blanket based feature subset selection algorithm that considers features both separately and together as a group in order to provide not just the best set of features but also the best combination of features

VeFloodBN: 基于离散贝叶斯网络的意大利威尼托大区市镇级洪涝风险建模。模型融合人均GDP、人口密度、坡度、河网密度、降雨频率等 12 个节点，构建因果有向无环图（DAG）；采用最大似然估计学习条件概率表，通过结构校验、马尔可夫毯与活跃路径分析保证模型合理性。支持精确变量消元，并对比拒绝采样、似然加权与吉布斯 MCMC 等近似推理方法。连续变量经分位数离散化后，实现逐市镇批量后验概率推断，利用 GeoPandas 输出空间风险热力图。针对 Top‑N 风险排名引入分层/比例采样策略，结合 Jaccard 与 Overlap 指标评估结论稳健性；同时提供单因子边际效应、稀有证据下采样效率对比及条件独立可视化，兼具可解释性与推断鲁棒性。

Code accompanying paper "Model-Augmented Conditional Mutual Information Estimation for Feature Selection" in UAI 2020

Personal Markov blanket and simple knowledge manager using Wikipedia. Recommends connected frontier concepts, sorts by in and out connection degree.

A K4-engine cybernetic runtime for AI swarms. Replaces semantic theater with thermodynamic constraint. Uses the K4 Algebra of Four-Fold Distinction to achieve phase-lock, prevent trajectory loss, and execute structurally coherent agent instantiation.

Multipath derivation of a formal algebraic framework — the Algebra of Four-Fold Distinction — which is then applied across physics, cognition, and AI alignment. Includes technical and subjective narratives.

This repository contains a collection of lab exercises and exams from the TDDE15: Advanced Machine Learning course taking at Linköping Univerity during the fall 2024. The main topics are: Bayesian Networks, Hidden Markov Models, Q-learning, REINFORCE, and Gaussian Processes.

William Blake, Professor Jiang, and the Architecture of Intelligence — rapid AI-augmented scholarly response to Game Theory #24: The AI Apocalypse (12 May 2026). Triangulates Jiang's diagnosis with Blake's perceptual diagnostics and Friston's Active Inference.

Cognition layer for autonomous agents — modular OSS packages for active-inference reasoning, thoughtseed competition, computational phenomenology, and workspace dynamics. Each adopts alone. Companion to elume (memory layer).

An audit-grade collaborative review of active inference free energy minimization, by an organic operator and AI co-contributors. Reviewable foundations, reproducible tests (87 deterministic pytest assertions across 11 numerical demonstrations), open tensions.

Assigments done as part of coursework

14+ candidate semantic dualities across 6 algebraic layers, 72 primitives connected by a DAG, validated via 8 domains + 3 negative controls (DVS), Bayesian model comparison (BF=121), and a 72-bit triadic head over GPT-2 Medium / GPT-Neo