Research collaboration
École polytechnique
Institut Polytechnique de Paris
I am Orso Forghieri, a researcher in reinforcement learning and applied mathematics. I study state abstraction, dynamic programming, and scalable algorithms for large sequential decision problems.
Since February 2026, I have been collaborating with École polytechnique on adaptive learning methods. My current work includes AggQL, a model-free RL method that refines its state representation using confidence-controlled temporal-difference residual tests, and Confidence-Adaptive Expert Disaggregation (CAED), a method for adapting large language models by dynamically creating LoRA specialists from statistically significant gradient disagreement.
My applied work connects these methods to resource allocation and latency-constrained edge-network optimization. I develop reusable Python libraries, benchmark algorithms, and build prototypes that validate decisions before integration.
Research
Reinforcement learning, state abstraction, hierarchical methods, and scalable dynamic programming.
Applied optimization
Resource allocation, service placement, and network routing under capacity and latency constraints.
Software engineering
Modular Python tooling, reproducible benchmarks, automated tests, and validated integration interfaces.
Selected software
Constrained optimization
service-placement-benchmarks
Research benchmark
Compare edge-network service placement and routing strategies under compute, link-capacity, and latency constraints.
Sequential decision-making
mdpforge
Python research library
Build finite Markov decision processes and compare solvers at a shared target accuracy with reproducible experiments.
Software integration
service-placement-deployable
Research / engineering prototype
Train PPO policies for edge-network service placement, routing, and bandwidth allocation, then validate and export decisions through CLI and HTTP interfaces.
Research interests
State abstraction and Markov aggregation/disaggregation; approximate dynamic programming; hierarchical reinforcement learning; planning and stochastic optimization for large MDPs; adaptive representations and parameter-efficient LLM adaptation.
Academic background
My PhD thesis, Hierarchical Reinforcement Learning for Large Scale Problems, was prepared at École polytechnique / Institut Polytechnique de Paris, within CMAP, under the supervision of Erwan Le Pennec, Hind Castel-Taleb, and Emmanuel Hyon.
I hold a Master’s degree from École normale supérieure Paris-Saclay and an Engineering Diploma from École polytechnique. My applied experience includes quantitative research at Qube Research & Technologies and service-placement research with Orange Gardens / CMAP. See my CV for experience and education, and Teaching for academic service.
Selected publications
- Hierarchical Reinforcement Learning for Large Scale Problems — PhD thesis, 2025.
- Faster Latency Constrained Service Placement in Edge Computing with Deep Reinforcement Learning — TX4NETs workshop at IFIP Networking, 2025.
- State Abstraction Discovery from Progressive Disaggregation Methods — EWRL, 2024.
- Progressive State Space Disaggregation for Infinite Horizon Dynamic Programming — ICAPS, 2024.
Contact
I welcome research and engineering discussions on scalable decision-making, reinforcement learning, and constrained resource optimization. Contact me at orso.forghieri@gmail.com or via the profiles below.