Ethan Decker Email & Phone Number

Electrical Engineer in circuit/hardware design and embedded programming.

-made improvements to time complexity of codebase for the calculation of a jacobian concerning the ith layer
-made memory complexity improvements by using forward differentiation and batched processing on certain inputs and outputs of neural network. I then, using theses techniques split the jacobian matrix calculation into vector jacobian product calculations such that we avoided a memory blowup mid-calculation.
-calculated the theoretical lower bound of flops for jacobian calculation of ith layer with respect to input to see if there can be further optimizations.
-created a program that has any neural networks python file as an input and creates a computational graph of the primal operations as output
-coded a general flops calculator for any defined neural network that outputs flops used during inference with the input of the neural network code and input.
-Derived the flops count on all cifar library models.
- Derived all flops count equations for the backpropagation of any neural network
-encoded all forward and back propagation flops counting objects into the codebase via a computational graph generator
-During my derivation, I spotted an inefficiency in the PyTorch codebase
-Actively making contributions to PyTorch execution engine codebase (C++) (with my professor help who was an ex-researcher at meta)

I am researching ways to describe nearly decoherence-free quantum systems with my professor.

- verified theorem of decoherence-free subspaces in a quantum computer system
- performed asymptotic analysis on a composite quantum system to gather the analytical effect of quantum decoherence over time.
-perturbed a composite system hamiltonian to relax a theoretical condition of decoherence-free subspaces to gain a model for decoherence over an arbitrary time
- Derived the consequence of this relaxation by expanding the system as a function of epsilon and solving
-Traced out the environment to show our added impurity does remain in the system.
-Generated practical example systems to apply equations to
-Generalized theory to 2 dimensional subspace to be applicable in a quantum computer setting
- calculated the fidelity between our initial density matrix and our density matrix at a later point in time to analyze the decoherence effects on our system
- wrote an interpretation and verified this result
-expanding the system to a second order series of epsilon to gain insight into decoherence, first order is still decoherence free
-calculated the fidelity of the second-order expansion, which shows a loss of information proportional to time squared.
-expanding to N dimensions to represent an arbitrary-sized Hilbert space.

Seattle, Washington, United States

Worked on the first ever online upload portal for consumer insurance documents at liberty mutual. -Improved first MVP design for iteration -chose a generator for the code base of the inherited MVP -upgraded all AWS dependencies to v2 -upgraded wrappers of CDK stacks to fit liberty mutuals internal network requirements -Assisted in the onboarding of two engineers by introducing them to various problem statements. -introduced new UI components for the first MVP-Modified old UI components of the code inherited

-Managed multiple labs of 100-Mentored 200 students through basic concepts of digital design -Guided students through the design of a fuller adder

Pittsburgh, Pennsylvania, United States

-generated an analytical equation for quantifying the free energy transfer of hydrophobic molecules-leveraged python for a approximate free energy crossover quantification -developed graphing software through python to handle high data inputs -leveraged mongoDB and pymongo for a high volume data storage pipeline

Doctor Of Philosophy, Computer Science

Bachelor'S Degree, Computer Science, 3.5

Activities and Societies: Active entrepreneur in the Technology Sector

Bachelor'S Degree, Physics, 3.5

Minor, Computer Engineering, 3.5

Minor, Cybersecurity, 3.5

Minor, Mathematics, 3.5