Robert Streit Email & Phone Number

Austin, Tx, Us

Austin, Texas Area

• Examining the design of parallel and distributed, or otherwise scalable, combinatorial optimization methods.• Contributed first sublinear complexity parallel method for computing optimal basis of a binary matroid.• Via rapidly mixing Markov chains, solving many problems – submodular minimization, max cut, and uniquecoverage – in parallel and universal manner by using algebraic and geometric representations.• Constructing distributed algorithms for solving semidefinite programs with complexity varying in sparsity.• Using the calculus of variations to analyze a user alignment in Bayesian multi-armed bandit recommender systems.

San Francisco, California, United States

• Engineered Python library for data driven optimization like Metropolis-Hastings walks, conjugate gradientdescent, quasi-Newton methods, and geometric methods and validate on real world dynamic pricing data.• Used contributions to develop risk factor estimation and model calibration program for user pricing ML models.• Collaborated with ML infrastructure team to standardize optimization in data pipelines across company.• Evaluated ML models in online stochastic pricing experiments with SQL and contributed sampling bias correction method.

Greater Seattle Area

• Created a robust system for streaming security analyses from unreliable data sources in Elastic stack for evaluation.• Increased security team’s positive influence by tooling the translation of data-driven security insights into accessible developer facing reports, contextualized by automatically generated recommendations and fixes for projects.

• Designed deep learning network hardware accelerator using data locality and hardware parallelization.• Compared approach to more traditional parallel linear algebra processors like systolic arrays.• Wrote Linux kernel drivers and modified PyTorch tools to use accelerator on Xilinx Zynq SoC running Ubuntu.• Led senior design team by assigning tasks, tracking milestones, compiling reports, and liaising with professors.

Austin, Texas Area

• Invented new fast automatic synchronization algorithm for parallel computing using incremental computation.• Researched competing algorithms and improved on their shortcomings.• Analyzed complexity and proved lower bounds on problem to show algorithm has tight worst case complexity.• Formalized custom language extending C++ and modified Clang compiler to support new paradigms.• Supplemented with empirical work, and compiled findings and commentary into undergraduate honors thesis.

Austin, Texas Area

• Developed tool used to visualize and understand decisions made by deep learning neural networks.• Researched algebraic methods for decomposing AI systems and evaluated their usefulness to company needs. • Implemented methods such as gradient based class activation mapping and deconvolutional networks.

Austin, Texas Area

• Made debugging & performance test tool interacting with 2000 devices concurrently in a network• Amended network drivers to support the parallel redundancy protocol in QNX-RTOS

Doctor Of Philosophy - Phd, Electrical And Computer Engineering

Bachelor’S Degree, Biomedical/Medical Engineering

Activities and Societies: IEEE Computer Society, UT ECE Undergraduate Advisory Board, Eta Kappa Nu IEEE Honor Society