research

Tensor Factorization and Regularized Optimization to Detect Latent Features in fMRI data

Final Project for Mathematics of High Dimensional Data (Graduate Level)

We present a scalable model to represent multi-subject, multi-dataset fMRI data in the same learned latent space. Firstly, we cluster similar subjects based on the tensor decomposition of the input data. Then, we use regularized gradient descent on the data from similar subjects to learn the shared latent space. Our best model consistently achieves accuracies over 85%. Moreover, our chosen regularization ensure scalability and low computational complexity.

Final Paper

Majority Dynamics and Information Aggregation in Social Networks

Final Project for Advanced Algorithm Design (Graduate Level)

We explore time-evolving graphs, where each node's 'opinion' is updated at every time-step to match the majority opinion held by its neighbors. After several time-steps, a winning opinion is declared by performing a majority vote on the population. We provide theory which shows that certain graphs eventually hold a majority incorrect opinion, despite each node being initially biased towards the correct opinion. Furthermore, we empirically investigate the effects of various seeding and opinion-recovery methods.

Final Paper

Transimpedance Amplifiers for High-Speed Fiber Optic Communication

Final Project for Wireless and High-Speed Integrated Circuits and Systems (Graduate Level)

We design a low-noise low-power transimpedence amplifier with a high bandwidth of 7 GHz and gain of over 4.3 kΩ. The design was analytically analyzed using the generalized time-constant method and simulated using Cadence.

Final Paper

Low-Cost Mechanical Ventilator for the COVID-19 Pandemic

Final Project for Robotics and Autonomous Systems

We present a design for a portable mechanical ventilator offering both volume and pressure controlled support, accessible through a clean user-interface. We construct sensors and valves from common hardware and 3D-printed parts, significantly reducing the cost-of-build without sacrificing safety. Our final design costs under $300, which is about one-tenth the cost of off-the-shelf solutions.

Final Paper Pitch Deck

Error Mitigation for Near-Term Quantum Computing

IBM Qiskit Camp and Hackathon

We implemented the Richardson extrapolation technique to reduce gate-errors in noisy intermediate-scale quantum computers, as proposed by Kandala, et. al. 2019 (Nature 567, 491–495).

Presentation Code