Leo Torres, PhD
Lead Data Scientist & Network Science Researcher
About
I am a Lead Data Scientist at FGS Global, building LLM-powered data pipelines for Fortune 500 clients while maintaining active research in network science. My work bridges cutting-edge academic research with large-scale industrial applications, processing millions of articles daily through advanced AI systems.
I earned my PhD in Network Science from Northeastern University under Tina Eliassi-Rad, and completed postdoctoral research at the Max Planck Institute for Mathematics in the Sciences in Jürgen Jost's Complex Systems group, developing novel spectral methods for graph analysis.
My research advances the mathematical foundations of network science through spectral graph theory, with applications ranging from theoretical analysis to production machine learning systems. I have published in leading venues including Physical Review E, SIAM Review, and Journal of Complex Networks.
Research Interests
Research & Industry Applications
I develop rigorous mathematical frameworks for understanding complex networks, applying spectral methods to both theoretical problems and large-scale industrial systems processing millions of data points daily.
Current Work & Research Program
Applied Network Science in Industry (2023-Present)
As Lead Data Scientist at FGS Global, I apply network science and spectral methods to large-scale text analysis and information retrieval systems. This work bridges my theoretical research with practical applications processing millions of documents daily for Fortune 500 clients.
Key technical contributions include: (i) LLM-powered pipeline architecture handling 10M+ documents daily, (ii) RAG systems with vector databases for domain-specific retrieval, (iii) Graph-based document clustering and similarity algorithms at production scale.
Spectral Theory of Non-Backtracking Operators
My primary research program develops the spectral theory of non-backtracking operators on graphs. Unlike traditional graph matrices (adjacency, Laplacian), the non-backtracking matrix is non-normal, requiring novel analytical techniques from functional analysis and operator theory.
Key contributions include: (i) Complete characterization of the spectrum for regular graphs, (ii) Perturbation theory for eigenvalues under graph modifications, (iii) Applications to network immunization and epidemic control.
Geometric Methods in Network Analysis
I develop geometric approaches to network analysis that leverage differential geometry and algebraic topology. This research program connects the local geometry of graph embeddings to global network properties.
Recent work includes the GLEE (Geometric Laplacian Eigenmap Embedding) algorithm, which preserves geometric structure in dimensionality reduction, achieving state-of-the-art performance on link prediction tasks.
Mathematical Epidemiology on Networks
During the COVID-19 pandemic, I applied network science methods to understand disease spread on mobility networks. This work combined rigorous mathematical modeling with large-scale data analysis of human movement patterns.
Our findings informed public health policy by quantifying the impact of mobility restrictions on disease transmission in major US metropolitan areas.