I've been getting into academic research slowly over the past couple months. A lot of my prior work is either under an NDA or something too niche to show off.
Approximate Attention
Work in Progress. A method to accelerate the computation of transformer attention by employing a hybrid between an approximation and exact evaluation of the softmax function.
Artificial Agents for Humanitarian Assistance
Technical Report. An exploration of the possible role of grounded LLM systems for humanitarian crisis through an agent designed for laws, regulations, and programs for refugees and asylum in the US.
Towards the Decentralized and Collaborative Training and Inference of Large Models
Whitepaper. A theoretical mathematical model for distributing computation of LLMs across untrusted (decentralized) peers by optimizations that maintain efficiency while remaining scalable.
Classifying Intellectual Property for Enterprise Data Security
Not Published. A system that employed fine-tuned and grounded LLMs for classifying whether outgoing data was intellectual property. Presented as an oral at the 2023 Southern California Conference for Undergraduate Research.