Omid Sadeghi

Omid Sadeghi

Lead Research Scientist, Optimization @ 84.51°

Optimization, decision science & machine learning

Biography

I am a Lead Research Scientist in Optimization at 84.51°, Kroger’s data science and analytics company, in Chicago. I build optimization and simulation tools for high-impact operational decisions: a Bayesian-optimization framework for labor scheduling across Kroger stores, a facility-location simulation that lets executives explore what-if scenarios for distribution-center network design, and a multi-objective model for weekly store-delivery scheduling that balances warehouse, store, and transportation costs.

Before joining 84.51° in September 2024, I was a Postdoctoral Associate at the MIT Sloan School of Management with Prof. Negin Golrezaei, working on incentive-compatible mechanisms for online marketplaces, auction design, and real-time algorithmic decision-making. I received my Ph.D. in Electrical and Computer Engineering from the University of Washington in December 2023, advised by Prof. Maryam Fazel; my dissertation, The Diminishing Returns (DR) Property and Its Applications in Machine Learning, focused on online and submodular optimization. Along the way I also earned an M.S. in Mathematics (Optimization) at UW. I hold a B.S. in Electrical Engineering and Mathematics from the Sharif University of Technology, and in the summer of 2015 I was a junior research assistant at the Chinese University of Hong Kong with Prof. Chandra Nair, working on invex functions and capacity bounds for very weak interference channels.

Skills

Core expertise
  • Bayesian optimization
  • Stochastic optimization
  • Convex optimization
  • Multi-objective optimization
  • Simulation
  • Online learning
  • Resource allocation
  • Mathematical modeling
Research areas
  • Sequential decision-making
  • Submodular optimization
  • Mechanism design
  • Privacy-preserving optimization
Tools
  • Python
  • PySpark
  • Scikit-learn
  • Pandas
  • NumPy
  • SciPy
  • CVXPY
  • Git
  • LaTeX
Languages
  • English (fluent)
  • Farsi (native)
  • Spanish (intermediate)
  • German (elementary)

In my spare time, you can find me learning languages (Spanish and German these days), going for a run, or playing soccer. I am also a certified Heroic Life Coach.

Research Interest

My work sits where optimization theory meets operational decisions. In industry, I turn Bayesian, stochastic, and multi-objective optimization into decision tools that planners and executives use. In my academic research, I designed algorithms with provable guarantees for non-convex problems in machine learning under real-world limits such as budgets, privacy, incentive compatibility, and fairness, in both online and offline settings, with applications to online advertising and resource allocation.

Current work at 84.51°

Labor scheduling

A Bayesian-optimization framework for labor-scheduling decisions across Kroger retail locations, balancing labor cost against operational performance.

Distribution-network design

A facility-location simulation and decision-support tool that lets executives evaluate what-if scenarios for distribution-center site selection and operational planning.

Store-delivery scheduling

A multi-objective optimization model for weekly store deliveries that jointly optimizes warehouse, store, and transportation costs under operational constraints.

Academic research

Most of my published work falls into one of these areas:

Online Learning Submodular Optimization Privacy and Incentives in ML

Publications

Peer-reviewed papers and workshop talks. Filter by topic, or see everything on Google Scholar.

INFORMS JOO 2024

Function Design for Improved Competitive Ratio in Online Resource Allocation with Procurement Costs

with Mitas Ray, Maryam Fazel, and Lillian J. Ratliff

INFORMS Journal on Optimization, 2024

TL;DRDevelops optimized bidding functions for online advertising and resource allocation when there are costs to acquire resources, improving competitive performance guarantees.

UAI 2024

Efficient Interactive Maximization of BP and Weakly Submodular Objectives

with Adhyyan Narang, Lillian J. Ratliff, Maryam Fazel, and Jeff Bilmes

UAI 2024

TL;DRCreates algorithms for optimizing complex objective functions with human feedback, useful for machine learning applications where the goal isn't perfectly defined upfront.

NeurIPS 2023

No-Regret Online Prediction with Strategic Experts

with Maryam Fazel

NeurIPS 2023

TL;DRDevelops prediction algorithms that work even when experts can strategically manipulate their advice, ensuring robust performance in adversarial settings.

ACDA 2023

Fast First-Order Methods for Monotone Strongly DR-Submodular Maximization

with Maryam Fazel

ACDA 2023

TL;DRCreates faster optimization algorithms for a special class of problems common in machine learning, achieving better theoretical guarantees and practical performance.

AISTATS 2021

Differentially Private Monotone Submodular Maximization Under Matroid and Knapsack Constraints

with Maryam Fazel

AISTATS 2021

TL;DRSolves optimization problems while protecting individual privacy, applicable to recommendation systems and data analysis where user privacy is critical.

ICML Workshop 2021

Improved Regret Bounds for Online Submodular Maximization

with Maryam Fazel

ICML 2021 Workshop on Subset Selection in Machine Learning: From Theory to Applications

TL;DRProvides better theoretical guarantees for online learning algorithms that select diverse and representative subsets of data in real-time applications.

AAAI 2021

Online DR-Submodular Maximization: Minimizing Regret and Constraint Violation

with Prasanna Raut and Maryam Fazel

AAAI 2021

TL;DRBalances optimization performance with constraint satisfaction in online settings, useful for real-time resource allocation with budget or capacity limits.

NeurIPS 2020 Spotlight

A Single Recipe for Online Submodular Maximization with Adversarial or Stochastic Constraints

with Prasanna Raut and Maryam Fazel

NeurIPS 2020 · Spotlight presentation (280/9454 submissions)

TL;DRCreates a unified algorithm that works optimally under both predictable and unpredictable conditions, eliminating the need for separate algorithms for different scenarios.

AISTATS 2020

Online Continuous DR-Submodular Maximization with Long-Term Budget Constraints

with Maryam Fazel

AISTATS 2020

TL;DRExtends optimization algorithms to handle continuous decisions over time while managing long-term budget constraints, applicable to online advertising and resource management.

ICML Workshop 2020

Online Algorithms for Budget-Constrained DR-Submodular Maximization

with Reza Eghbali and Maryam Fazel

ICML 2020 Workshop on Negative Dependence and Submodularity for ML

TL;DRDevelops online optimization algorithms that maximize objective functions while staying within budget limits, with applications to advertising and resource allocation.

Notes and Surveys

Course projects and surveys from my time at UW.

PDF Spring 2021

Data-Dependent Regret Bounds for Bandits Problems

with Max Gray and Tanner Fiez

TL;DRSurvey of multi-armed bandit algorithms with performance guarantees that adapt to the actual difficulty of the problem instance, providing tighter theoretical bounds.

PDF Winter 2021

Online Adversarial Zero-Sum Games

TL;DRComprehensive notes on game theory algorithms for competitive scenarios where one player's gain equals another's loss, with applications to security and economics.

PDF Spring 2020

Introduction to Spectral Graph Theory

with Catherine Babecki and Kevin Liu

TL;DRTutorial on using linear algebra to analyze network structures, covering eigenvalues and eigenvectors of graphs with applications to clustering and network analysis.

PDF Winter 2018

Linear Regression and Sequential Experimental Design

with Johannes Linder, Felix Leeb, and Sumit Mukherjee

TL;DRGuide to optimally designing experiments over time to learn linear relationships, balancing exploration of new conditions with exploitation of known information.

Wisdom Unlocked

The three big ideas I took away from each book I've read. Pick a cover to open it.

3 big ideas

    Contact

    The best way to reach me is by email. I'm always happy to chat about optimization, online learning, or good books.

    Google Scholar Papers and citations LinkedIn in/sadeghiomid X (Twitter) @omidsUW