cv

PhD

University of Michigan, Ann Arbor, MI

Aerospace Engineering and Scientific Computing

• Large Language Models
• Numerical Linear Algebra
• Methods and Practice of Scientific Computing

MSE

University of Michigan, Ann Arbor, MI

Mechanical Engineering

• Model Predictive Control
• Computational Data Science and Machine Learning
• Inference, Estimation, and Learning

BSc

Bogazici University, Istanbul, Turkey

Mechanical Engineering

MICDE 2023-2024 Graduate Fellow

Michigan Institute of Computational Discovery in Engineerings

Best Reproducibility Award

Michigan Institute of Data Science

2023 MICDE Annual Symposium Poster Competition runner up

Michigan Institute of Computational Discovery in Engineerings

Graduate Student Research Assistant

Department of Aerospace Engineering, University of Michigan

• Leading a cross-university research team of graduate and undergraduate students to develop a framework for behavioral cloning using multi-modal data
• Developed and implemented tensor decomposition algorithms for scientific machine learning
• Mentored 2 master’s students and 1 undergraduate student in research setting
• Presented research findings at 5+ peer reviewed papers and 10+ international conferences

Generative Modeling Using Tensor-Network Embeddings

• Working on creating an architecture to generate new gameplay sequences using tensor network embeddings of ATARI games.
• Studying state-of-the-art generative models for video data and comparing them with tensor network-based approaches.

Bayesian Optimal Experimental Design in Tensor-Network Reduced Spaces

• Developed a framework for large-scale, high-dimensional data using tensor decompositions
• Enhanced measurement accuracy by up to 18% through optimal sensor placement

Incremental Hierarchical Tucker Decomposition

• Developed the first incremental algorithm for hierarchical Tucker decomposition in the literature
• Achieved up to 60% reduction in computational cost compared to existing methods
• Authored a manuscript detailing the algorithm for peer-reviewed journal submission
• Implemented the algorithm as high-performance scientific computing software

Incremental Tensor Train Decomposition

• Developed a state-of-the-art algorithm for converting tensor streams into tensor train format
• Reduced computation time by 95% and increased compression ratio by 57× compared to existing methods
• Achieved up to 13x speedup in end-to-end training time for deep learning against AE/VAE based architectures
• Released the algorithm as an open-source software package at github.com/dorukaks/TT-ICE

Neural Network Inverse Design for Self-Oscillating Gels

• Designed a neural network to predict physical and motion parameters of a PDE-driven chaotic system
• Achieved over 99% accuracy for discrete parameters and 98% for continuous parameters

Process Parameter Control for Fused Deposition Modeling

• Engineered and built a cost-effective bead height measurement system for Ultimaker 3D printers
• Established a model linking process parameters to bead cross-sectional geometry
• Demonstrated up to 85% reduction in bead height error through experimental testing
• Presented findings at the 2020 American Control Conference

Incremental Hierarchical Tucker Decomposition

arXiv (submitted to JMLR)

An Incremental Tensor Train Decomposition Algorithm

SIAM Journal on Scientific Computing (SISC)

Low-Rank Tensor Network Encodings for video-to-action behavioral cloning

Transactions on Machine Learning Research (TMLR)