Independent Researcher — France
Email: ulrich.tan@polytechnique.org
ORCID: https://orcid.org/0009-0001-2907-501X
About me
I work on continuous learning, synaptic plasticity, and geometric learning rules, with the goal of understanding how biological principles can inspire more adaptive, efficient, and robust AI systems.
My research focuses on how local, biologically plausible learning rules—such as dendritic integration, nonlinear synaptic subunits, and structural plasticity—can be translated into geometric and dynamical frameworks for machine learning. I am particularly interested in how these mechanisms can support lifelong learning, sparsity, stability, and energy‑efficient computation.
This line of work has led me to explore:
- dendritic segment integration and nonlinear synaptic subunits,
- Wasserstein geometry as a foundation for Hebbian learning,
- geometry‑driven structural plasticity and emergent pruning,
- learning dynamics under noise and spontaneous activity,
- resource‑efficient and edge‑deployable learning mechanisms.
Beyond theoretical work, I also build AI systems and engineering prototypes, including:
- lightweight continuous‑learning frameworks
- dendritic computation models
- edge‑AI experiments
- distributed identity and cryptographic systems (DLT, ZKP, MPC, PQC)
- GPU‑accelerated pipelines and LLM‑based architectures
My broader interest is in how adaptive, decentralized, and resource‑efficient learning can shape the next generation of AI systems—especially in contexts where compute, energy, or privacy constraints matter.
Research Interests
- Continuous Learning & Lifelong Adaptation
- Synaptic Plasticity & Dendritic Computation
- Learning Dynamics & Geometric Methods
- Optimal Transport & Wasserstein Geometry
- Neuro‑Inspired Machine Learning
- Edge AI & Resource‑Efficient Learning
- Emergent Structure & Sparsity
Publications
A Wasserstein Geometric Framework for Hebbian Plasticity
Ulrich Tan
arXiv:2604.16052
A geometric formulation of Hebbian plasticity based on W₂ geodesics, showing that classical Hebbian updates can be interpreted as contractive flows in Wasserstein space.
X announcement :
I’ve just released a new preprint proposing a geometric framework for memory and learning dynamics using Optimal Transport.
— Ulrich Tan (@Ulrichxtan) April 24, 2026
Geometry‑Driven Structural Plasticity Induces Synaptic Pruning Under White Noise
Ulrich Tan
(2026, WIP)
Shows that a Wasserstein‑based structural update rule (HWG) produces spontaneous pruning, branch‑level competition, and symmetry breaking under pure white noise — unlike classical STDP or non‑geometric structural plasticity.
👉 PDF (preprint) — to be added soon
Related Projects
Dendritic Segment Model
A computational model of nonlinear dendritic integration and synaptic subunits.
👉 dendritic_segment_model
SPEANN
A lightweight experimental framework exploring synaptic plasticity and continuous learning mechanisms.
👉 speann
Other Work
I also maintain a portfolio of engineering projects (AI systems, distributed identity, data engineering, OS development):
👉 portfolio
AI Systems & Engineering Work
In parallel to theoretical research, I design and build production‑grade AI systems, including:
- sovereign LLM platforms (secure cloud, GPU clusters, LLMOps)
- RAG and agentic AI architectures
- evaluation and monitoring pipelines
- privacy‑preserving systems (DLT, ZKP, MPC, PQC)
- distributed data platforms and analytics engines
- edge‑AI prototypes and resource‑efficient models
This dual perspective—biological inspiration + large‑scale AI engineering—shapes my approach to innovation and emerging technology.
Contact
For collaboration, research discussions, or AI systems work, you can reach me at:
Email: ulrich.tan@polytechnique.org
LinkedIn: https://www.linkedin.com/in/ulrich-t-1b83064
GitHub: https://github.com/jietra