Yuyang Ma

Yuyang (Johnny) Ma

Ph.D. student at USC

I am a Ph.D. student in Industrial and Systems Engineering at the University of Southern California, advised by Dr. Karmel S. Shehadeh. I develop optimization models and algorithms for decisions under uncertainty, with applications in logistics, mobility, and energy systems. My recent work includes integrated planning of drone-based disaster relief (accepted at Transportation Science) and deployment, repositioning, and pricing of mobile EV-charging fleets under demand learned from context and price.

I hold an M.S. in Operations Research from Georgia Tech and bachelor's degrees in Industrial Engineering from the University of Pittsburgh and Sichuan University.

Open to internships & part-time research

Looking for Summer 2027 research / applied scientist internships (and part-time research during the year) in logistics, marketplace, and mobility teams. I study optimization under uncertainty and connect demand prediction with operational decisions. Expected graduation: May 2028.

Optimization

Robust · Distributionally robust · Stochastic

Algorithms

Mixed-integer programming · C&CG · K-adaptability

Machine learning

Demand modeling · Time-series forecasting · Deep RL (in progress)

Tools

Python · Gurobi · PyTorch · SQL

News

Oct 2026

Our paper on integrated drone-based disaster-relief planning was accepted by Transportation Science.

Nov 2025

Our paper "Integrated Planning of Drone-Based Disaster Relief: Facility Location, Inventory Prepositioning, and Fleet Operations under Uncertainty" can be now accessed on SSRN 🥳.

Aug 2025

I am now continuing my Ph.D. journey at University of Southern California.

Dec 2022

I am honored to obtain Master's degree from Georgia Tech.

Selected Works

Integrated Planning of Drone-Based Disaster Relief: Facility Location, Inventory Prepositioning, and Fleet Operations under Uncertainty

Published

Yuyang Ma, Karmel S. Shehadeh, Man Yiu Tsang

PDF SSRN

Distributionally Robust Deployment, Repositioning, and Pricing of Autonomous Mobile Charging Stations under Context- and Price-Dependent Demand

Working paper

Yuyang Ma, Karmel S. Shehadeh

Teaching

Lehigh University

Georgia Institute of Technology

Contact

Feel free to reach out about collaborations, talks, research opportunities, or anything else you’d like to discuss. Email is the best way to reach me (yuyangm 'at' usc 'dot' edu); please include a brief subject line, and I’ll respond as soon as I can.

Office Address

3650 McClintock Ave, Room 304
Los Angeles, CA 90089

Visitor Map

Explore visitor map & stats

Visitor map by MapMyVisitors.

Download PDF

Miscellaneous Moments

When I am not solving optimization problems, I enjoy exploring nature, photography, and spending time with my cats.

Xiaobao 2024

Accepted · Transportation Science

Integrated Planning of Drone-Based Disaster Relief: Facility Location, Inventory Prepositioning, and Fleet Operations under Uncertainty

Authors: Yuyang Ma, Karmel S. Shehadeh, Man Yiu Tsang Status: Accepted by Transportation Science · October 14, 2026

Motivation

When earthquakes and hurricanes strike, roads collapse and communities become isolated overnight—yet those first hours are when aid matters most. From Japan's Noto Peninsula earthquake to Hurricane Dorian in the Bahamas, destroyed infrastructure has left survivors without food, water, or medical care for days. Drones offer a lifeline: they fly over rubble and flooded highways to deliver critical supplies directly to those in need. But planning drone relief is far from simple—decisions about facility placement, inventory, fleet size, and drone assignments are deeply coupled and must be made under profound uncertainty about disaster outcomes.

Illustration of drone-based disaster relief operations showing four phases: pre-disaster planning, disaster disruption, post-disaster without reassignment, and post-disaster with reassignment
Drone-based disaster relief: (a) pre-disaster planning, (b) disaster disruption, (c) post-disaster without reassignment, and (d) post-disaster with reassignment.

What We Do

We introduce a two-stage robust optimization framework that tackles all of these decisions simultaneously. Before the disaster, the model determines where to build relief facilities, what to stockpile, how many drones to deploy, and where to assign them. After the disaster, it adaptively distributes supplies and—crucially—reassigns surviving drones to the areas that need them most. A key innovation is our decision-dependent uncertainty set, which captures, for the first time, that a drone's post-disaster functionality depends on which facility it was assigned to before the disaster.

Why It Matters

Existing models fix drone assignments before the disaster and never revisit them—if a drone is destroyed, its assigned communities go without aid. Our model breaks this rigidity through post-disaster drone reassignment. Computational case studies based on the 2013 Lushan and 2008 Wenchuan earthquakes in China show that allowing reassignment can reduce unmet demand and operating costs under disruption. The benefits vary by disruption scenario. With our approach, drones held in reserve at intact facilities are dynamically redeployed to cover the gaps—turning potential shortages into delivered aid.

Key Contributions

  • An integrated robust optimization model for facility location, inventory prepositioning, fleet sizing, and drone assignment under decision-dependent uncertainty
  • A novel post-disaster drone reassignment mechanism that provides decision-makers with contingency plans, dramatically reducing unmet demand
  • A K-adaptability approximation solved via a tailored column-and-constraint generation algorithm
  • Validation on two real-world earthquake case studies demonstrating substantial operational benefits

Resources

Download PDF View on SSRN
Working paper · 2026

Distributionally Robust Deployment, Repositioning, and Pricing of Autonomous Mobile Charging Stations under Context- and Price-Dependent Demand

Authors: Yuyang Ma, Karmel S. Shehadeh Status: In preparation

Motivation

Autonomous mobile charging stations (AMCSs) bring the charger to the parked EV. They are useful where fixed chargers are unavailable, congested, or hard to install—shopping centers, airports, campuses, event venues. A provider operating a fleet of AMCSs faces a familiar marketplace problem: where to position supply, what to charge, and how to adjust during the day as demand reveals itself. Demand depends on price and on context such as weather, traffic, and events, and it is only partially predictable.

What We Do

We study a two-period daily planning problem. Before the first period, the provider allocates the fleet across service areas and sets prices. After observing first-period demand and updated context, it can reposition vehicles and re-price for the second period. Demand in each period is modeled with scenario-specific demand functions estimated from historical records: contextual features identify the operating scenario, demand responds to the area's own price, and the second-period forecast carries over the first-period forecast residual. Rather than trusting a point forecast, we hedge against forecast error with mean–MAD ambiguity sets whose means depend on the pricing decision itself.

The resulting model is a three-stage distributionally robust optimization problem with integer repositioning decisions and decision-dependent ambiguity. We reduce it to a two-stage robust problem, approximate second-period pricing with K candidate pricing policies (K-adaptability) selected after first-period demand is observed, show that the repositioning recourse is equivalent to a linear program through a minimum-cost-flow structure, and solve the problem with a column-and-constraint generation algorithm.

Why It Matters

The framework connects demand learning with operational decisions in a way that is practical for operators: a small menu of pricing plans prepared in advance, fully adaptive repositioning, and protection against the forecast being wrong. The same structure applies to other on-demand fleets—ride-hailing, delivery, and shared mobility—where supply must be positioned and priced before demand is known.

Status

Modeling and solution methodology are complete; numerical experiments are in progress. A preprint will be posted here when available.