Optimization modeling

Optimization problems appear in industrial applications from manufacturing and distribution to healthcare. However, most such problems are still solved heuristically by hand rather than optimally by state-of-the-art solvers, as the expertise required to formulate and solve these problems limits the widespread adoption of optimization tools and techniques. My work has developed several tools to simplify optimization modeling. Most recently, we released a Large Language Model (LLM)-based agent to formulate and solve MILP problems from natural language descriptions.

Talks

• Big Data is Low Rank using LowRankModels(keynote at JuliaCon, June 2019) video

• The Type of Language for Mathematical Programming (JuliaCon, June 2017) slides github video

Software

• OptiMUS: model optimization problems using natural language

• Linnaeus: detect equivalence between optimization algorithms

• Convex.jl: disciplined convex programming in Julia

Papers

OptiMUS: Optimization Modeling Using MIP Solvers and Large Language Models
A. AhmadiTeshnizi, W. Gao, and M. Udell
Submitted to ICLR, 2023
[arxiv][url][bib]

An automatic system to detect equivalence between iterative algorithms
S. Zhao, L. Lessard, and M. Udell
2021
[arxiv][pdf][url][slides][bib]

Disciplined Multi-Convex Programming
X. Shen, S. Diamond, M. Udell, Y. Gu, and S. Boyd
Chinese Control and Decision Conference (CCDC), 2017
Best Student Paper
[arxiv][bib]

Convex Optimization in Julia
M. Udell, K. Mohan, D. Zeng, J. Hong, S. Diamond, and S. Boyd
SC14 Workshop on High Performance Technical Computing in Dynamic Languages, 2014
[arxiv][code][bib]