Hi! I'm Jasmine Shone, a current student at MIT. This summer, I was a SWE Intern at Meta working on a computer vision model launch at Meta Superintelligence Labs and large-scale load testing at Instagram.
I enjoy learning about and exploring lots of different fields. Within AI that means Computer Vision, NLP, medical ML, deep learning, and robotics. Outside of AI that means data science, full-stack development, infrastructure, game development, and quantitative finance. In my free time, I enjoy singing/composing music, as well as writing short stories and personal essays.
For a full list of my work experiences, please see my LinkedIn profile.
Meta Superintelligence Labs: Working on a new vision model launch.
Instagram: Working on large-scale load testing infrastructure.
Core (SWE): Completed five C++ projects focused on building infrastructure for live trading systems. Gained hands-on experience with performance optimization techniques, including efficient data structures and memory management. Actively participated in code reviews to improve code quality and robustness.
Algo Development: Designed, programmed, and deployed a live trading bot for Brazilian equities that achieved net positive returns in production. Applied data science techniques to train predictive models, evaluate performance metrics, and effectively communicate insights to stakeholders.
Making robots generalize across varying object poses, camera views, and object instances with only 10 demonstrations.
I'm currently interested in world-modelling and 3D generation. Relatedly, I'm interested in retrieval, memory, and context— making models that remember and see beyond immediate inputs/tasks. I'm also interested in intelligence which comes from compression, with feature/representation learning being a subset of that interest. Within representation learning, I've been curious recently about vision model representations and cross-modality information loss.
Make sure to check out the launch!
We extend the I-Con framework to discover new losses which achieve state-of-the-art results on clustering and supervised learning.
We create a new paradigm of image-pair similarity learning conditioned on text descriptions.
By utilizing priors from Vision-language models and image features from large classification models to create a novel keypoint abstraction for robot actions, we generalize effectively across object poses, camera viewpoints, and object instances with only 10 demonstrations.
Are fluctuations in gas prices predictive of unemployment rates and how do regional differences in mass transit and gasoline production affect this trend?
We introduce SketchAgent, a novel framework for generating sequential sketches from language prompts. By leveraging large language models and diffusion models, we achieve high-quality sketch generation that captures the essence of the input text.
We create an LLM in-context learning pipeline to systematically optimize (1) maximum token length of the prompt, (2) the mechanism of choosing few-shot examples, and (3) the ordering of few-shot examples to generate applications.
We improve stable diffusion's ability to generate high-quality fundus images of the eye, specifically for glaucoma, by finetuning on an extremely small dataset of 170 images.