Publications
- Real-Time Vessel Segmentation for Ultra-Sound Guided Surgery
- Period: Feb 2024 - November 2024
- Status: Published as second author in High Performance Extreme Computing (HPEC ‘24). Supervised by MIT-LL
- Highlights:
- Investigated various pruning/quantization techniques for real-time image segmentation of human vessels.
- Contributed to writing a custom application for evaluating performance on a mobile AI system.
- Motivated the processing of tomographic segmentation algorithms from a discrete computer to a mobile AI system in the next generation of AI Guide.
- Large Reasoning Models for 3D Floorplanning in EDA
- Period: Aug 2023 - Ongoing
- Status: Under review, submitted as first author. Supervised by Qualcomm Fellowship and Prof. Franzon
- Highlights:
- Developed an auto-regressive decision-making model to enhance 3D IC floorplanning.
- Implemented a new architecture integrating sequence-to-sequence reinforcement learning, improving reasoning over large discrete action spaces.
- Achieved improvements in sample efficiency and floorplan quality through training with non-expert trajectories.
- Conducted comprehensive evaluations demonstrating advanced performance in wirelength reduction and multi-objective optimization.
- The Over-Certainty Phenomenon
- Period: Sept 2022 – Ongoing
- Status: Under review, submitted as first author, supervised by Prof. Jung-Eun Kim
- Highlights:
- Introduced a novel memory-efficient unsupervised domain adaptation algorithm (UDA) to improve calibration.
- Identified key issues in state-of-the-art UDA algorithms that harm model calibration.
- Retained comparable accuracy to SOTA.
- Can Low-Rank Knowledge Distillation be Useful for Microelectronic Reasoning?
- Period: March 2024 - May 2024
- Status: Published as co-first author, LLM-Aided Design (LAD ‘24)
- Highlights:
- Presented empirical results on the feasibility of using offline LLMs in EDA.
- Evaluated Llama-2-7B’s performance as a microelectronics Q&A expert.
- Introduced a novel LLM adaptation technique, low-rank knowledge distillation (LoRA-KD).
- Released an evaluation benchmark to support future research.
- Optimal Brain Dissection
- Period: May 2022 – Aug 2023
- Status: Published as first author in BioInspired Processing (BIP ‘23), supervised by Sozzani Lab and USDA
- Highlights:
- Won Best Paper award.
- Introduced a SOTA technique for feature-importance determination.
- Developed the dense autoencoder, a new architecture for reducing reconstruction error in -omics data.
- Outperformed the de facto gene regulatory network in explaining gene expressions.
- DepthGraphNet
- Period: Oct 2022 – July 2023
- Status: Published as first author in Machine Learning for Computer Aided Design (MLCAD ‘23)
- Highlights:
- Investigated the use of siamese-graph neural networks for circuit graph isomorphism (CGI) detection.
- Demonstrated logarithmic run-time complexity with respect to graph size.
- Outperformed all other classical and neural methods in CGI detection accuracy.
- Introduced theorems for the optimal architecture of GNNs for CGI detection.
- Network Inference Approach for Phosphoproteomics
- Period: May 2022 – Nov 2022
- Status: Published as second author in Methods in Molecular Biology (MIMB vol. 2690), supervised by Sozzani Lab
- Highlights:
- Described methods to statistically analyze label-free phosphoproteomic data and infer post-transcriptional regulatory networks over time.
- Used the Bayesian Dirichlet Equivalent Uniform to inference underlying latent relationships between variables.
Note: During my time at NC State, I wanted to enrich my knowledge beyond what I was learning in class and within Prof. Franzon’s lab. To do this, I joined two additional labs: the low-resource computing lab under Prof. Jung-Eun Kim in the CS department and Sozzani lab under Prof. Ross Sozzani in the Microbial Biology department. I concurrently produced research for all three labs while balancing my responsibilities as a student, junior advisor, teaching assistant, and research/grant proposal writer. I have manuscripts being prepared; contact me for details.