As quantum devices realize more and more physical qubits, they are still limited to running circuits of one and two qubit gates. While this is sufficient for universality, it can lead to circuits of high depth which are subject to more errors. Multi-qubit gates would be able to significantly reduce the circuit depth. However, multi-qubit gates are subject to a much larger control space than smaller gates. This increased complexity is a fundamental barrier to quantum control optimization (choosing how the gate is realized). The space is complicated and too large to sweep.

This is where the Reinforcement Learning approach is effective. When the control parameter space grows much too large for a team to manually sweep, the RL-based gate design can identify effective controls. There is also work shown to indicate that this approach is robust and that the RL models are effective at control optimization.

Our work uses gradient-based reinforcement learning techniques such as supervised pretraining to warm start the unsupervised learning process of methods like Proximal Policy Optimization (PPO), to tailor policy solutions towards the goal on hand. This allows us to to improve upon various previously known control solutions while focusing on reducing leakage, gate times and gate infidelities.

This project is an ongoing collaboration with PhD student Adyant Kamdar and Prof. Murphy Niu. This project was also awarded 10,000 GPU Hours by the NVIDIA Academic Grants Program

This project was part of the Research in Industrial Projects (RIPS) Research Experience for Undergrads (REU) hosted at the Institute for Pure and Applied Mathematics (IPAM) at UCLA. Our Industry sponsor was the Aerospace Corporation.

A Lunar Time Estimation system would be an essential part of any future prolonged Lunar missions or even Lunar Surface Networks. A Lunar Time Estimation system is a way of keeping precise and synchronized time estimates across a network of Lunar base stations. Bhamidipati et al. (2022) propose a modified Diffusion Kalman Filter for Lunar timing, which uses intermittently available Earth-GPS signals and measurements from neighboring stations to maintain accurate, synchronized time estimates.

A Kalman filter is a statistical algorithm which balances measurements and estimates for a more accurate estimation than the two could give individually. A Diffusion Kalman Filter is a network of Kalman Filters which use each other's estimates as measurements which factor into their own estimates. My group implemented the Diffusion Kalman Filter proposed by Bhamidipati et al. Then, we optimized our filter for the weight of the neighboring estimates in personal estimation. We evaluate and compare the performance of a range of cost functions and optimization techniques, with applications to optimization on generalized diffusion algorithms. 

We also studied the proposed Diffusion filter from a statistical perspective, most notably deriving the bias of our filter's estimate. Showing that it has non-zero bias, unlike the traditional Kalman Filter.

Wildfires stand as an ever present dangerous natural phenomenon, especially in California. However, their recent rise warrants extra attention. In particular, when fires start it’s important to understand how they spread. This project explores the relationship between the spread of invasive weed species across California and the regions with frequent fires. 

To perform our analysis we used linear algebra tools such as projection, linear regression, kNN and PCA, we utilized geological data of wildfires and records of weed coverage to correlate the spread of weeds and the frequency of fires. 

Although this data doesn't imply certain weeds cause more/larger fires we can analyze what can make a plant more likely to have been in a fire. Some notable contributing factors can include disturbed land, climate conditions, or biome changes (Ex. dried swamps). Using some comparisons to data found online as well as in other research on invasive grasses we can largely confirm our conclusions for several species, such as, notably, Ricinus communis (castor bean), Festuca myuros (rat-tail fescue) , and Verbascum thapsus (great mullein). Overall it is evident that certain weeds are a greater risk than others, even among invasive species as a whole.