Luke McDermott

Hi. I'm a second year ML PhD student in UCSD's ECE Department, co-advised by Rahul Parhi & Robert W. Heath. Previously, I was a research scientist at Modern Intelligence and completed my undergrad at UCSD in Math-CS.

I am interested in developing efficient deep learning models, incorporating sparsity and plasticity to create adaptive, lifelong AI. Currently, I am researching parametric forms of attention which learn at test time.

Selected Work

LoLA: Low Rank Linear Attention with Sparse Caching
by Luke McDermott, Robert W. Heath, & Rahul Parhi.

Linear Mode Connectivity in Sparse Neural Networks
by Luke McDermott & Daniel Cummings.
Accepted in NeurIPS 2023's UniReps Workshop.

UniCat: Crafting a Stronger Fusion Baseline for Multimodal Re-Identification
by J. Crawford, H. Yin, L. McDermott, D. Cummings.
Accepted at NeurIPS 2023's UniReps Workshop.

Contact

• lmcdermo@ucsd.edu
• Google Scholar
• X/Twitter
• Linkedin

Feel free to reach out for questions or collaborations.