I develop machine learning and statistical methods for high-dimensional, structured, and incomplete data — spanning applied ML, uncertainty quantification, scientific computing, and healthcare analytics. I care about models that are not only accurate but reliable enough to support real decisions.

I am a Ph.D. Candidate in Industrial & Systems Engineering at Georgia Tech, advised by Professor Kamran Paynabar and Professor Xiao Liu. I hold an M.S. in Statistics from Georgia Tech and an M.S. in Industrial Engineering with a focus on Healthcare Systems from Amirkabir University of Technology.

Open to Data Science, Machine Learning and Applied Scientist internships and full-time opportunities — across technology, healthcare and other data-intensive applications. Get in touch.

Featured Work

  • Hospital capacity for medical tourists and local patients

    A hybrid agent-based and discrete-event simulation of a multi-specialty international-patient department, with patient behaviour — abandonment, doctor switching, channel choice — modelled explicitly. Against an otherwise identical discrete-event-only model it cuts medical tourists' hospital-queue wait by 82% (13.7 to 2.4 days), and compatible-section bed sharing cuts the inpatient admission queue by 43% (25.9 to 14.8 days, 30 randomized-seed replications, p < 0.001).

    arrivals queue Ward A Ward B idle beds shared mean admission queue 25.9 → 14.8 days · 43% shorter

    Healthcare operationsSimulationAnyLogicDesign of experiments

    Paper Code DOI

  • Constrained boosting for reduced-order models

    Predicting a simulation's reduced basis at unseen parameters means regressing a point on a manifold, valid only inside a bounded region. cXGBoost enforces that bound inside every leaf of the boosting ensemble rather than correcting afterwards, so predictions are valid by construction and no general-purpose optimiser is called at any split.

    valid unconstrained leaf value outside the valid region projected into both balls so the prediction is always usable enforced at every split, during fit not corrected afterwards

    Gradient boostingManifold learningSurrogate modeling

    Paper Code

  • Learning from delayed feedback

    When an action's outcome only arrives after a delay, most algorithms wait or discard the signal. State-EXP3 charges one delayed outcome to every action that could have produced it, and pays for that pooling with an effective dimension that reads the overlap between actions rather than counting them — so the cost tracks how much the actions actually share.

    actions a₁ a₂ a₃ states s₁ s₂ delay d loss at s₁ credited to a₁ and a₂, the actions that reach s₁ price of pooling effective dimension vₜ set by how much actions overlap

    Sequential decision makingBanditsOnline learningDelayed feedback

    Paper Code

  • Deciding when partial clinical evidence is enough

    Clinical models often have to act on incomplete inputs — imaging but no labs, notes but no genomics. This work certifies a prediction when the evidence already in hand determines the decision, and otherwise abstains and says which modality to acquire next, with distribution-free coverage guarantees.

    observed imaging labs notes genomics enough evidence? yes certify & act no abstain, acquire next coverage holds for every stopping pattern

    Conformal predictionMultimodal learningMissing dataClinical decision support

    Paper Code

  • Surrogates that keep the data's shape

    Simulation output is naturally multiway — space by space by time. The usual move is to flatten it into one long vector, which throws that structure away. This keeps it: every mode gets its own low-dimensional subspace, each interpolated on its own manifold, with the cores aligned before a single low-rank surrogate is fitted. Across four PDE benchmarks that is 35.6× to 564× less interpolation work and 4× to 108× faster than flattened baselines, matching their accuracy to within 0.35 percentage points.

    simulation field space × space × time one subspace per mode U₁ U₂ U₃ core each mode interpolated on its own manifold cores aligned, then one low-rank surrogate vs flattening the field up to 564× less work 4×–108× faster, same accuracy

    Tensor decompositionsSurrogate modelingRiemannian geometryScientific ML

    Manuscript under review

Technical Skills

Programming & data
PythonSQLRPySparkNumPypandasSciPyMATLAB
Machine learning
scikit-learnXGBoostPyTorchDeep learningRepresentation learningMultimodal learningDiffusion models
Statistics & modeling
Statistical modelingNonparametric statisticsGaussian processesDesign of experimentsUncertainty quantificationCalibration
Optimization & simulation
Mixed-integer programmingAnyLogic (ABS + DES)ArenaSimulinkMonte CarloGAMS
Tools & computing
GitLinuxAWSCloud computingGPU / HPCSLURMLaTeX
Research methods
Conformal predictionBanditsTensor methodsReduced-order modelingCFDFEALLM-based information extraction

Experience

  • 2023 – Present

    Graduate Research Assistant — Georgia Institute of Technology

    H. Milton Stewart School of Industrial & Systems Engineering · Advisors: Kamran Paynabar, Xiao Liu

    Develop machine learning and statistical methods for high-dimensional data — gradient boosting, deep learning, uncertainty quantification, multimodal learning and surrogate modeling. Design and evaluate end-to-end ML pipelines in Python, PyTorch and scikit-learn across healthcare, scientific computing and simulation datasets, and carry projects from problem formulation through experimentation, statistical validation, reproducible code and publication.

  • 2022 – 2023

    Course Instructor — Artificial Intelligence Laboratory

    Royan Institute, Tehran

    Designed and delivered a project-based AI curriculum covering rule-based expert systems, hybrid architectures, and applied machine learning in clinical healthcare contexts. Supervised student projects from problem formulation through evaluation.

  • 2021 – 2023

    Graduate Research Assistant

    Amirkabir University of Technology, Tehran · Prof. Hadi Mosadegh

    Healthcare systems engineering: built a hybrid agent-based and discrete-event simulation in AnyLogic of a hospital serving medical tourists and local patients, modelling patient behaviour, resource sharing, admissions and capacity constraints. Designed and evaluated a compatible-section bed-sharing policy: medical tourists' hospital-queue wait fell 82% and the inpatient admission queue 43% against a discrete-event-only baseline. Quantified patient flow, abandonment, emergency escalation and resource utilisation.

  • 2018 – 2019

    Research Intern — R&D Division

    Arvand Pishro (Automotive R&D), Safadasht, Iran

    Designed and validated a closed-loop fuel tank cooling and monitoring system integrating sensor feedback and automated temperature control.

  • 2018 – 2020

    Undergraduate Research Assistant — Machine Learning & Simulation

    Amirkabir University of Technology, Tehran · Prof. Mehdi Sabzeh Parvar

    Built a physics-based simulator in MATLAB/Simulink from system velocity equations and operational constraints, generated training and test data across operating scenarios, and implemented machine learning models in Python — including a three-layer neural network — to predict system behaviour and expand the feasible operating range under hybrid inputs.

Projects

  • Predictive maintenance and anomaly detection — end-to-end pipeline for sensor-based fault detection: ARIMA and LSTM forecasting, ensemble classifiers over multivariate telemetry, and hybrid Weibull–state-space degradation modelling with online retraining, giving over 15 hours of advance fault warning. Deployed on Spark for low-latency inference across sensor networks. Python, PySpark · 2025
  • Multimodal phenotype modeling — built representation-learning pipelines over single-cell transcriptomics, longitudinal EHR and imaging-derived phenotypes, using contrastive learning for patient and cellular embeddings and transformer language models to extract structured phenotypes from unstructured clinical notes. Python, PyTorch · 2024–25
  • Drug safety and clinical risk — detected adverse drug events from longitudinal EHR with temporally-aware patient embeddings over medication trajectories, labs and comorbidities; compared regularized regression, tree ensembles and neural models, and audited calibration and fairness across demographic strata. Python · 2024–25
  • Sensitivity-guided diffusion models — constrained diffusion generative models to physically meaningful latent subspaces, deriving sensitivity-informed perturbation schedules that steer the forward SDE along high-variance directions to improve sample efficiency in data-scarce scientific regimes. Python, PyTorch · 2024–25
  • Anomaly detection in sensor streams — built ARMA-based statistical monitoring with sliding-window control charts to flag faults in industrial sensor streams early, trading off detection delay against false-alarm rate. Python · 2023
  • Loan default prediction — trained and compared ensemble classifiers for consumer credit risk, evaluating ranking performance alongside probability calibration rather than accuracy alone. Python · 2023
  • Safe operating range under hybrid inputs — built a six-degree-of-freedom physics-based simulator in MATLAB/Simulink from system velocity equations and operational constraints, generated data across operating scenarios, and trained neural network models in Python to predict system behaviour and expand the feasible operating range. MATLAB, Simulink, Python · 2018–20
Additional projects
  • Internet access and female economic empowerment — cross-national nonparametric analysis of internet usage against female employment and representation in senior management across 92 countries (World Bank Gender Data Portal, 2022); Mann–Whitney U tests with Bonferroni correction and Anderson–Darling / Shapiro–Wilk normality diagnostics. Group project
  • Breast cancer prediction — built and statistically evaluated diagnostic classification pipelines, comparing generalization across models rather than training accuracy. Python, scikit-learn · 2021–22
  • Nurse scheduling — formulated and solved mixed-integer programs for healthcare workforce scheduling under demand uncertainty. GAMS · 2020–21
  • Healthcare operations quality — applied Six Sigma to locate and reduce service quality deficiencies in a healthcare organization. 2021–22
  • Food recommender expert system — knowledge-based recommendation tailored to individual health characteristics and dietary needs. Python · 2021–22
  • Tensor-structured learning pipelines — deep and tensor-structured architectures for parameterized scientific fields under geometric constraints. Python, PyTorch · 2024
  • Transonic airfoil CFD — numerical analysis of transonic flow over RAE airfoils and nonlinear PDE solution behaviour under varying physical parameters. Ansys Fluent · 2021–22

Publications

Full list on Google Scholar.

Healthcare analytics

  1. Reducing Waiting Time for Medical Tourists through Hybrid Agent-Based and Discrete-Event Simulation: A Hospital Case Study

    Melika Baghi, Hadi Mosadegh

    Under review, Health Care Management Science · arXiv:2605.19139

    Paper Code DOI

  2. Guaranteed Adaptive Modality Acquisition: When the Policy Chooses Its Own Calibration Group

    Melika Baghi

    Under review · arXiv:2608.15520

    Paper Code

Machine learning methods

  1. Constrained Extreme Gradient Boosting for Adapting Reduced-Order Models

    Melika Baghi, Xiao Liu, Kamran Paynabar

    arXiv:2605.04130, 2026

    Paper Code

  2. Pooling and Drift in Delayed Bandits

    Melika Baghi

    ML×OR 2026 Workshop (non-archival) · arXiv:2609.01761

    Paper Code

  3. Supervised Active-Subspace Input Reduction and Risk-Controlled Field Uncertainty for Tensor-Valued Computer Experiments

    Melika Baghi, Xiao Liu, Kamran Paynabar

    Manuscript under review

  4. Tensor-based Reduced Order Model

    Melika Baghi, Xiao Liu, Kamran Paynabar

    Manuscript under review

  5. Reference-Measure Conformal Prediction for Subspace-Valued Outputs

    Manuscript in preparation

  6. Topological Conformal Risk Control: Distribution-Free Uncertainty for Persistent Features

    Manuscript in preparation

Education

  • 2023 – Present

    Ph.D., Industrial & Systems Engineering — System Informatics & Control

    Georgia Institute of Technology · Advisors: Kamran Paynabar, Xiao Liu

    Dissertation: Geometry-Aware Learning Frameworks for Reduced-Order Modeling with Uncertainty Quantification in High-Dimensional PDE Systems.

  • 2023 – 2026

    M.S., Statistics

    Georgia Institute of Technology

  • 2021 – 2023

    M.S., Industrial Engineering — Healthcare Systems

    Amirkabir University of Technology · GPA 4.0/4.0 · Advisor: Hadi Mosadegh

    Thesis on scheduling and patient flow for medical tourists. Direct entry as an elite student, ranked 3rd of 120 applicants.

  • 2018 – 2021

    B.S., Industrial Engineering

    Amirkabir University of Technology · GPA 3.91/4.0

  • 2016 – 2020

    B.S., Aerospace Engineering

    Amirkabir University of Technology · GPA 3.79/4.0

    Selected as an elite student and approved to pursue both degrees concurrently.

Teaching

  • 2024 – Present

    Teaching Assistant, then Head Teaching Assistant — ISyE 6525: High-Dimensional Data Analytics

    Georgia Institute of Technology

    Lead a TA team for a graduate course of 150+ students per semester on dimensionality reduction, sparse modeling, manifold learning, and high-dimensional regression. Named Outstanding Online TA in 2025.

  • 2022 – 2023

    Teaching Assistant — Principles of Simulation

    Industrial Engineering, Amirkabir University of Technology · Prof. Abbas Ahmadi

    Instruction in Python, MATLAB, Simulink, Arena and AnyLogic for stochastic simulation and computational modelling, including healthcare operations simulation.

  • 2020 – 2023

    Teaching Assistant — Engineering Dynamics

    Aerospace Engineering, Amirkabir University of Technology · Prof. Hamid Reza Ovesy

    Weekly recitations on analytical mechanics, Lagrangian dynamics and stability analysis. Ranked #1 teaching assistant in the department by student evaluation for three consecutive years.

Honors & Awards

  • 2026Most Innovative Approach Award, NSF Future Manufacturing Data Challenge
  • 2026Third-year Progress Travel Award, H. Milton Stewart School of Industrial & Systems Engineering
  • 2025Travel Award, Institute for Mathematical and Statistical Innovation (IMSI) — an NSF-funded institute
  • 2025Outstanding Online TA Award, ISyE 6525 — Georgia Institute of Technology
  • 2023Stewart Topper Fellowship, Georgia Institute of Technology
  • 2023Mansour Doctoral Fellowship, North Carolina State University