MELMO (Moreau Envelope Smoothing with Linear Minimization Oracles)Research
2026 — A non-Euclidean optimizer for non-smooth composite optimization problems, currently under review at a top-tier journal. It leverages Linear Minimization Oracles (LMOs) to compute geometry-aware descent directions that accelerate convergence. While offering competitive theoretical bounds, it usually outperforms existing methods in empirical evaluations.
RL for vehicle routing problems with emission quotaResearch
2025 — A hybrid approach combining operations research heuristics with reinforcement learning for Vehicle Routing Problems under emission quotas. A discrete-event simulator benchmarks the hybrid OR+RL method against classical approaches. Results demonstrate that the hybrid approach outperforms traditional techniques while achieving significantly faster computation times.