Welcome to my personal GitHub repository! This space serves as a central hub for my research, prototypes, and creative explorations across AI, economics, optimization, and systems design.
I’m Arthur Maffre. I come from a background in economics, where I developed a deep interest in analytical thinking and the logic of decision-making under uncertainty. For me, rationality isn’t about rigidity — it’s about understanding the world in a way that respects individual choices and freedoms. I enjoy engaging in open discussions that challenge ideas and reveal why, behind seemingly abstract models, economics is ultimately about people, choices, and trade-offs. Today, my research bridges Generative Flow Networks (GFlowNets), bilevel optimization, and dynamic portfolio management, with a focus on making machine learning models more interpretable and decision-oriented.
I’m passionate about:
- 🧠 AI for structured decision-making
- 📈 Quantitative finance & portfolio generation
- 🧩 Game theory & economic optimization
- 🧮 Mathematical modeling in complex systems
- 🚀 Long-term vision: impactful research and real-world applications
This repository includes:
- 📜 High-level project descriptions and links to submodules
- 🧪 Research notebooks and proof-of-concept implementations
- 🧠 Drafts, visualizations, and experimental insights
- 🛠️ Utilities, snippets, and productivity tools
- 🌱 Evolving frameworks around GFlowNet optimization and simulation
Projects here are often in active development — many are stepping stones toward publishable work or larger initiatives.
| Project | Description |
|---|---|
GFlowNet-LLM-Bayes |
The core thesis: true AI is not just next-token prediction, it’s structured belief revision under uncertainty. This project frames an adversarial game where a GFlowNet generates sequences that break Bayesian consistency, forcing the LLM to restore coherence. The aim: an LLM that internalizes causal schemas and updates beliefs like a true Bayesian reasoner. |
GFlowNet–Bilevel–Knapsack |
Uses a GFlowNet with critic to estimate Z and accelerate solving a bilevel knapsack problem with Benders cuts. |
GFlowNet-Knapsack-CDF |
A GFlowNet that learns probabilistic solutions to the 0‑1 Knapsack problem, enabling efficient global optimization. |
Transformer-Portfolio |
Transformer-powered GFlowNet generating sequences of portfolio allocations with Sharpe ratio optimization. |
BayesianAxioms-GFN |
GFlowNet-based inference of rational choice axioms within a Bayesian structure learning framework. |
streamlit-dashboards |
Elegant and interactive visualizations used for hackathons and research presentations. |
| Project | Description |
|---|---|
RMBP-finance |
A project to democratize access to advanced AI-driven investing — with just two simple sliders (risk and ESG preferences), anyone can align their portfolio with both financial goals and personal values. |
I aim to develop tools and models that map the solution space — not just maximize a function. My goal is to build interpretable, robust, and scalable solutions to problems where economics, AI, and complexity intersect.
To organize my ideas and research projects, I use a 5-level system that helps track progress from early exploration to polished work:
| Level | Description |
|---|---|
| 1 | 🌱 Idea stage — rough notes, spontaneous insights, and unstructured thoughts. |
| 2 | 🧪 Exploratory phase — initial experiments, toy models, and feasibility checks. |
| 3 | 🔧 Prototyping — focused implementations, small-scale validation, and initial write-ups. |
| 4 | 📈 Refinement — robust models, formal results, and preparation for dissemination (talks, reports). |
| 5 | 🚀 Publication-ready — polished papers, arXiv submissions, or open-source releases. |
This system helps me prioritize, iterate, and communicate progress effectively across projects.
Feel free to reach out or follow my work:
- 📧 [email protected]
- 🌐 [Coming soon] Personal website
All content in this repository is shared under the MIT License, unless otherwise specified.
“Build to explore. Explore to understand. Understand to transform.”
— Arthur