Ludovic Morin

About

Cornell master’s student (CS/Stats) focused on research-heavy ML projects: implemented a sparsity-driven RL approach (DAPD + Soft Actor-Critic) that maintained stable learning even with ~95% of weights removed in OpenAI Gym continuous-control tasks. Also worked on diffusion-based computer vision with conditioning and latency-focused U-Net choices, and scaled unsupervised community detection on a 50k-node/800k-edge Reddit graph via BFS subgraph sampling.

Experience

Education

Key Strengths

• Built an end-to-end RL research system integrating DAPD into Soft Actor-Critic
• Empirical stability analysis across sparsity levels; found policies remained stable with up to ~95% weights removed
• Methodical debugging of training collapse/oscillation via repeated trials and gradient behavior analysis
• Pragmatic scaling decision on large graph data (50k nodes, 800k edges): replaced adjacency-matrix approach with BFS subgraph sampling and rewrote training loop
• Applied efficiency tradeoffs in generative modeling (lighter-weight U-Nets to reduce diffusion model latency)
• Production-minded ML perspective emphasizing latency, compute constraints, safety margins, staged rollout, and reproducibility