The MIT SmartSolve project seeks to advance AI‑guided algorithmic discovery and accelerate computations by generating improved strategies for algorithm and architecture selection. Our current work targets challenges in computational linear algebra, addressing the increasing complexity of choosing efficient solvers, data formats, precision settings, and hardware resources for structurally diverse matrices—an area where conventional approaches leave significant room for improvement.

Methodology

Our methodology involves building a comprehensive performance database through systematic benchmarking and applying automated Pareto analysis to reveal optimal trade‑offs between accuracy and speed. This database serves as the foundation for a data‑driven model that synthesizes dispatch strategies tailored for high‑performance linear algebra software.

Publications

• Outstanding Short Paper Award. Rushil Shah, Emmanuel Lujan, Rabab Alomairy, and Alan Edelman. “Data-Driven Dynamic Algorithm Dispatch with Large Language Models,“ 2025 IEEE High Performance Extreme Computing Conference (HPEC) (link).
• Emmanuel Lujan and Alan Edelman. “When Structure is Silent: Opportunities for Algorithmic Dispatch in Linear Algebra,” 2025 IEEE High Performance Extreme Computing Conference (HPEC) (link).
• Rabab Alomairy, Felipe Tome, Julian Samaroo, Alan Edelman. “Dynamic Task Scheduling with Data Dependency Awareness Using Julia”, 2024 IEEE High Performance Extreme Computing Conference (HPEC) (link).

Awards

Outstanding Short Paper Award — “Data-Driven Dynamic Algorithm Dispatch with Large Language Models”, IEEE High Performance Extreme Computing Conference (HPEC), 2025.

Talks

• Rushil Shah, Emmanuel Lujan, and Rabab Alomairy. “Automated Algorithm Selection Discovery via LLMs,” JuliaCon 2025, Lightning Talk. (Link).
• Alan Edelman et al. “Julia, Portable Numerical Linear Algebra and Beyond.” Presentation at Householder Symposium, 2025. (Link). Accessed June 20, 2025.
• Alan Edelman, “Improving the HPC Experience: Did Julia Get It Right, or Will AI Hide the Problem (or Both)?” Keynote at the Workshop on Asynchronous Many-Task Systems and Applications (WAMTA), 2025. (Link). Accessed June 20, 2025.

Software

Emmanuel Lujan, Rushil Shah, Rabab Alomairy, and Alan Edelman. “SmartSolve.jl: AI for Algorithmic Discovery, ” 0.1.0-alpha. Zenodo (link). All code and data used in this study are publicly available at the following GitHub repository. This repository contains the implementations, scripts, and instructions required to reproduce the results presented in this project.

Team

MIT’s Julia Lab: Accelerating Computation through a marriage of Computer Science & Computational Science.

Acknowledgements