View
More
 Hello!

I'm Ariel N. Lee

I am a UCLA alum and recently completed my MSc in Electrical & Computer Engineering at Boston University, focusing on machine learning and data analytics. My primary research has been in computer vision and SOTA LLMs, under the mentorship of Nataniel Ruiz. Specifically, I've been working on fine-tuning using LoRA modules to solve specific tasks, with a commitment to releasing high-quality open-source models and datasets. Future work on Mixture of Experts and specialized models!
Download Resume
Platypus: Quick, Cheap, and Powerful Refinement of LLMs

Ariel N. Lee, Cole Hunter, Nataniel Ruiz

arXiv preprint arxiv:2308.07317 (2023) | Models & Dataset: garage-bAInd | GitHub Repo

Our best model was the global leader in open-source SOTA LLMs at the time of writing. We release our entire dataset, fine-tuning and merging pipeline, and models to the research community.

Hardwiring ViT Patch Selectivity into CNNs using Patch Mixing

Ariel N. Lee, Sarah Adel Bargal, Janavi Kasera, Stan Sclaroff, Kate Saenko, Nataniel Ruiz

arXiv preprint arXiv:2306.17848 (2023) - Under Review

Released with this paper are two new datasets: Superimposed Masked Dataset & Realistic Occlusion Dataset

Meta AI Video Similarity Competition

8th overall (196 participants) | 1st in AI graduate course challenge (42 participants)

Used a pretrained, Self-Supervised Descriptor for Copy Detection model (ResNeXt101) to find similar, manipulated videos in a dataset of 40,000+ videos. 

Ensemble Effect: Leveraging Fine-tuned Models for Prompt Prediction

GitHub Repo

AI research project for predicting text prompts of generated images using an ensemble of multimodal models, including CLIP, BLIP, and ViT.

Custom, high-quality dataset of 100,000+ generated images, cleaned to have low semantic similarity.

Image prompts scraped from Midjourney discord channel

BU Wheelock Educational Policy Center: Analyzing Classroom Time

MLOps Development Team | Data & Process Engineer

Partnered with TeachForward and Wheelock Educational Policy Center to develop a feature extraction pipeline, analyzing the use of teaching time based on 10,000+ videos of classroom observations. Created a simple user interface for client using gradio and Hugging Face spaces.

Visual Odometry: Mapping Out the Camera Path

3rd in Computer Vision course challenge

GitHub repo

Task: Estimate a camera's path by tracking relative motion between successive frames, only using OpenCV for initial feature detection and matching.

Implemented RANSAC and linear triangulation from scratch for fundamental matrix and camera pose estimation, respectively.

Contact

Let's connect!

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.