you are interested in working with me, please follow the instructions at https://sites.usc.edu/dslab/prosepective/.
My current work includes the following:
1. COVID-19 – How to accurately forecast epidemics when dynamics
change due to changing policies? How to manage resources based on the
forecasts? How to come out of “stay-home” order?
involved: Machine learning, epidemiological models,
Publications: KDD 2020 (Health Day), Pre-print (not peer-reviewed), GitHub code, Web-interface; Contributing to
US forecast Hub, Germany+Poland
2. Deep learning-based prefetcher
– How to design highly compact deep learning models to predict future memory
accesses? How to utilize the predictions to design a prefetcher? How to
implement the prefetcher on hardware (FPGA/eFPGA)?
involved: Deep learning, meta-learning,
reinforcement learning, cache, FPGA
Safer connected communities – Using the full potential of
machine learning, social network analysis, and algorithms to improve society.
How to predict and evaluate crime? How to reduce violence among the homeless?
involved: Machine Learning, algorithms, social
ML-based Compiler for DSL to 1000 node systems
– For a given scientific computing program, what is the best way to decompose
it into tasks that potentially leads to least execution time? How to map tasks
to available resources (CPUs and GPUs in 1000s of nodes) to minimize execution
time? How to train such ML-based mappers when we do not have access to actual
resources during training? How to decide during runtime if the task mapping
needs to be changed? How to predict execution time of programs?
Skills involved: Deep learning, reinforcement learning,
meta-learning, integer linear programming approximations
5. Scaling Neural Networks
– How to compress neural networks that translate well in terms of increasing
throughput in hardware implementation? How to distribute training of neural
networks? How to design algorithms for faster training of neural networks?
Convolutional neural networks, model compression, algorithms, FPGA,
graph convolutional neural networks, parallel computing
Publications: HiPC 2019, FPGA
2020, ICLR 2020, IPDPS