Please find the details of a few research projects available at the University of California, Santa Cruz.
Request you to please go through the mail and Interested students may participate and benefit out of the opportunity.
Large-scale biological simulations of viruses:
The simulation of the self-assembly of a virus is an open problem in computational biology of tremendous importance, and solving it can help discover a new generation of antivirals. This project aims to use physics-based atomic simulations to simulate the self-assembly of a virus capsid. We leverage Molecular Dynamics simulations that are accelerated by Machine Learning models, and run these in a distributed fashion over large compute clusters. Strong skills in programming (PyTorch), ML, mathematics (e.g. differential equations) are required. Students will also be expected to delve deeper in physics (statistical mechanics), quantum chemistry and biology.
Differentiable simulators:
The dream of brain MRI imaging is to see very detailed tissue microstructure for diagnosing brain diseases or understanding the nature of intelligence. In order to push the frontier of brain MRI, one can use the strong inductive bias of MRI simulators, which simulate the scanning process of a brain phantom given a particular MRI sequence. We aim to build *differentiable* simulators for brain T1 MRI and diffusion MRI. While some simulators for MRI already exist which computationally simulate the scanning process of a brain phantom given a particular MRI sequence, we aim to make such simulators differentiable, in order to perform backpropagation through the simulator, just like with a neural network. Strong programming (PyTorch, Rust) and mathematical skills required (fourier analysis and differential equations in these particular projects).
ML compositionality:
We aim to create a graph of Machine Learning models that is both compositional and modular. Each node is a dataset and each directed edge v1 -> v2 is a Deep Learning model that takes input data v1 and outputs v2. The graph will be put on a public web platform and grown through crowdsourcing. Any new user will be able to commit (just like git commit) a new dataset (node) or a new model (edge linking two nodes). The compositional nature of the graph will enable entirely new tasks to be performed: any directed path connecting two nodes will become an ML predictor, data can be "flowed across" the entire graph and transformed to the representation of any of the nodes, and new datasets and modalities can be easily added over time. Strong skills in cloud computing, object-oriented programming, PyTorch and pipelining are required.
More details about these projects can be found on:
Prof Marinescu's website: https://razvanmarinescu.com. UC Santa Cruz expects students to be strongly motivated and be able to dedicate at least 20+ hours/week on the project. If the project outcomes are very good, University of California, Santa Cruz will consider funding for the students to present their work at international conferences, or to go to California for the summer to work with their research group, and even potential opportunities for PhD studies at UC Santa Cruz.
If you're interested, please get in touch with Prof. Marinescu at ramarine@ucsc.edu with your CV and a short description of previous work you've done.