Performing quantum compilation for the quantum fourier transform problem using a placement approach.
Implemented Google’s “A graph placement methodology for fast chip design” and improved their model design and problem formulation. Verified the correlation in placement quality between placing cell-clusters and placing cells themselves to justify the effectiveness of reinforcement-learning-based placement techniques in placement quality.
Proposed and implemented algorithms for non-integer-multiple-height (NIMH) standard cell placement for modern floor-plans.
Designed novel algorithms based on delayed hierarchical routing to handle incremental changes to the floor-plan.
Designed stacked-autoencoders that could filter out noises without supervision, improving signal-to-noise-ratio (SNR) by 2 fold.
Designed an alternative way of pre-training BERT that improves pre-training time and data efficiency by 10-times.
Proposed and verified HuBERT’s K-means and categorical loss function setting is equivalent to auto-encoders with permissive reconstruction losses in high dimensional vector spaces.
Institute of Health Policy and Management, Prof. Tzu-Bin Lu
Link to heading
Built a Markov chain to simulate numbers of people infected after being vaccinated and the cost-effectiveness with different vaccine coverage to help assess public health policies.