Eldar Kurtić

Focus on efficient inference with very large models

Developing methods for compression of deep neural networks. We make them smaller and faster, both on GPUs and CPUs.

We make deep neural networks run really fast on CPUs and GPUs

Computer vision for autonomous driving: - research and development of an ENet-like neural network capable of real-time semantic segmentation on compute-limited hardware (Python and PyTorch) - deployment of the neural network for use-cases on a real car (C++ and Robot Operating System ROS)

Worked on real-time trajectory planning for autonomous driving in highway scenarios: - lateral and longitudinal trajectories obtained by minimizing jerk (discomfort) via the Euler-Lagrange method for solving optimal control problems - motion prediction for other traffic participants and collision checks - implementation in C++ and integration with MATLAB/Simulink simulation environment

Responsible for teaching, designing, grading homework assignments and laboratory experiments. Courses: - Mathematics for Engineers 1 - Linear Algebra and Geometry - Fundamentals of Programming - Programming Techniques - Digital Electronics - Electronic Elements and Circuits - Power Electronics

Worked in the Research and Advance Engineering division on: - design, optimization, and automation of extrinsic camera calibration for an industrial robot arm in an automated optical inspection system; implementation in C# - programming Nextage Open dual-armed robot for joining tasks, establishing baselines for Reinforcement Learning experiments; implementation in Python and Robot Operating System (ROS)

Internet of Things (IoT) developer (team project): - design of a hardware board for the Ludo game - controller implemented in C - software enabling online games with remote players implemented in C#

Software developer (team project): - design and development of a single-page application for testing employees' technical skills - implementation with C#, AngularJS, .NET

https://micromasters.mit.edu/ds/