Logan Bolton
MSCS student at NYU Courant
Machine Learning Intern
@ Los Alamos National Lab
Interested in AI/ML and Computer Vision
About Me
- Graduated from Auburn University with a B.S. in Computer Science and a minor in Statistics.
- Beginning an M.S. in Computer Science at NYU Courant in Fall 2026.
I am currently looking for AI/ML related Summer 2027 internships. Feel free to reach out!
Research
Vision Language Models are Blind
Analyzed how and why VLMs such as GPT-4o have high scores on multimodal benchmarks, but often fail on extremely simple, abstract vision tasks.
This work has been featured by OpenAI, Google DeepMind, ByteDance, and has been cited 250+ times.
HoT: Highlighted Chain of Thought for Referencing Supporting Facts from Inputs
Explored how allowing LLMs to highlight their chain of thought to visually link information in the question and answer can increase LLM accuracy while also improving user experiences.
Understanding Generative AI Capabilities in Everyday Image Editing Tasks
Native multi-modal image-editing models like Nano Banana have shown an impressive ability to edit images through natural language prompts. This paper examines the strengths and weaknesses of these models when pitted against one-to-one matchups against human editors.
Experience
Machine Learning Intern
Los Alamos National Laboratory - Los Alamos, NMDesigning rotation-invariant neural network architectures for molecular simulations.
Undergraduate AI Researcher
Explainable AI Lab - Auburn UniversityFirst-authored three papers on understanding and improving multimodal LLMs/image-editing models in Dr. Anh Nguyen's Explainable AI lab.
Machine Learning Research Assistant
Sport Biomechanics Lab - Auburn UniversityCollaborated with an interdisciplinary team on a DoD-supported project studying how different movements affect the long-term health of canines. Trained pose estimation models to track joint positions across video data and analyzed kinematic patterns over time.
Machine Learning Intern
Corvid Technologies - Huntsville, ALDeveloped deep learning models to predict radar responses from 3D point clouds. Built roto-translation equivariant graph neural networks to leverage the geometric structure of point cloud data for more accurate signal prediction.
Projects
The Best Way to Make Your (Small) Vision Language Model Smarter
A blog post on using test-time compute to boost the accuracy of small VLMs.
Wordle - GRPO Agent
Finetuned an open-source LLM and trained with reinforcement learning to improve the model from only being able to complete 0.1% of Wordle games to successfully completing 18% of games.
Reverse Engineering Information from LLM Attention Values
For the final project of my graph theory class, I trained a model to reconstruct the adjacency matrix of a graph based off the attention values of an LLM, demonstrating how the attention patterns of LLMs can be used to help interpret their inner workings.
Contact
Have a question or want to work together? Email me at: loganbolton101@gmail.com