Computer Science PhD candidate at Harvard Insight + Interaction Lab led by Prof. Fernanda Viegas and Prof. Martin Wattenberg. 📖 I am interested in making generative AIs more controllable through mechanistic interpretability and internal world models.
Before entering Harvard, I was a Computer Science and Engineering student at Bucknell University 🦬 advised by Prof. Joshua Stough (Bucknell University) and Prof. Christopher Haggerty (Columbia University & New York-Presbyterian). My past projects focused on segmenting sparsely annotated medical videos using multi-task learning 🧡. My works were funded by the Ciffolillo Healthcare Technology Inventors Program (HTIP) 🏥. Our papers ( [1] [2] ) are published at the SPIE Medical Imaging 2021 & 2022 Conferences with oral presentations 📝.
I also developed a color analysis toolkit for films (KALMUS) 🎬. You can find the project's GitHub Repo here, KALMUS-Color-Toolkit. KALMUS' development was supported by the Mellon Academic Year Research Fellowship awarded 🥇 by Bucknell Humanities Center, and now used as a instructional software at Bucknell.
Reviewer for EMNLP, NeurIPS 2024 Creative AI Track, and NeurIPS Interpretable AI Workshop.
First-year/Pre-concentration Advisor for Harvard College.
Judge for National Collegiate Research Conference 2024 (NCRC) at Harvard.
Not all users are turned away by chatbots for making suspicious requests. Your identity plays a key role when chatbot decides whether to refuse your potentially problematic inquiry.
Our new study shows that the ChatGPT guardrails sensitive knowledge to users with different gender, racial, and political demographics. In particular, younger, female, and Asian-American users are more likely to trigger the refusal from the ChatGPT when querying sensitive information. We proposed a new evaluation framework to identify such bias in the chatbot's refusal behaviors.
Paper Preprint (to appear at EMNLP 2024 Main)
Have you ever thought about if chatbot LLMs are internally modeling your profile? If they are, how might this model of you influence the answers they give to your questions?
Our experiments provide evidence that a conversational AI is internally profling its user during the chat. We design an end-to-end prototype—TalkTuner—that expose this internal user model to the users. User study shows this new chatbot UI design impacts the user's trust in AIs and expose biases of LLM systems.
Paper Preprint
Recent work from Gurnee et al. show that the activations of neural language models have high correlation with the spatial and temporal properties of their inputs. However, the work didn't estabilish a causal link between two. Our experiments, filling a part of this blank, show that editing LLM's spatial representations can improve the model's performance on a simple spatial task.
Paper Preprint
Does the 2D image generative model has an internal model of 3D geometry? Can a 2D neural network see beyond the X-Y dimension of a matrix of pixels? Our project found controllable representations of 3D geometry inside diffusion model.
Paper
Youtube Video (55K Views)
Wonder how attention flows inside your Vision Transformer? What visual patterns are recognize by machine's attention? Does machine's attention resemble human's visual cognition? Collaboration with Catherine Yeh. My main contribution is the visualization of learnt attention in vision transformer models.
Paper
KALMUS is a Python package for the computational analysis of colors in films. It provides quantitative tools to study and compare the use of film color. This package serves two purposes: (1) various ways to measure, calculate and compare a film’s colors and (2) various ways to visualize a film’s color.
We aim to further improve the accuracy and clinical applicability of echocardiography video segmentation by extending the analysis from half-heartbeat (End-diastolic to End-systolic phases) to multi-heartbeat video. We proposed a sliding window data augmentation technique for efficiently learning moition tracking and semantic segmentation from sparesely annotated echo videos (only 2 annotations per video).
Paper .
We assessed a 3D-UNet's performance on jointly segmenting sparsely annotated half-heartbeat echocardiography videos and estimating cardiac structure's motions. The 3D-UNet was trained on CAMUS dataset, and we evaluated its generalizability on Stanford's EchoNet dataset. Comparing with traditional frame-based segmentation method, our results show that the joint learning of motion tracking and segmentation enhances the segmentation performance on video data. The video model also has better generalizability on unseen dataset than frame-based model.
Designing a Dashboard for Transparency and Control of Conversational AI, https://arxiv.org/abs/2406.07882
Demo GithubLinear probe found representations of scene attributes in a text-to-image diffusion model
Demo GithubFor more information, have a look at my curriculum vitae .
Served as the judge for National Collegiate Research Conference 2024 at Harvard. Reviewed and oversaw 8 undergraduate student projects.
I am now a new member of Insight + Interaction Lab at Harvard SEAS. It's so exciting to work with everyone here! Check out our group: Insight + Interaction Lab.
I graduated from Bucknell with Bachelor degree in Computer Science & Engineering and Summa Cum Laude distinction. I am pleased to receive the Bucknell Prize in Computer Science and Engineering (1 per class year), University Prize for Men, and President’s Award for Distinguished Academic Achievement.
In this fall, I will join the Harvard SEAS to pursue a Doctorate degree in Computer Science. It's my honor to be mentored by Prof. Fernanda Viegas and Prof. Martin Wattenberg ( Insight + Interaction Lab).
Our paper, "Fully automated multi-heartbeat echocardiography video segmentation and motion tracking", has been published at the SPIE Medical Imaging 2022: Image Processing Conference.
You can find the manuscript and recorded presentation here: link.
Our paper, "KALMUS: tools for color analysis of films", has been published on the Journal of Open Source Software! Our manuscript is open access, and you can find it here: link.
The GitHub repo of associated Python package, KALMUS, is here: link.
For installation instruction and detailed usage guide, please refer to the KALMUS documentation page.
Our paper, "Assessing the generalizability of temporally-coherent echocardiography video segmentation", has been published at SPIE Medical Imaging: Image processing 2021 conference!
You can find the manuscript here: link.
The presentation slides are available here: link.
Our paper, "Assessing the generalizability of temporally-coherent echocardiography video segmentation", has been accepted for an oral presentation by SPIE Medical Imaging: Image processing 2021 conference!
The preprint of paper is available here: link.
Thank you so much for visiting my website!