About me

I’m a fourth-year PhD student at UC San Diego, advised by Berk Ustun. My research focuses on making generative models and ML systems more reliable, interpretable, and human-aligned.

My work includes:

Augmented Generation Systems: retrieval-augmented generation (RAG) for LLMs, structured inductive biases for diffusion models
Human-interactive Systems: concept-bottleneck models (CBMs), privacy-preserving models for healthcare
Document Analysis and Security: adversarial robustness, graph-neural networks for document manipulation detection

Education

Harvard College: B.A. in Computer Science, 2019

UC San Diego: PhD in Computer Science and Engineering, 2021-present

Work Experience

Google Research: Student Researcher, 2024

Meta: PhD Research Intern, 2024

Twitter: Machine Learning Research Engineer, 2021

Lendbuzz (fintech startup): Machine Learning Research Engineer, 2019-2021

Publications

Sufficient Context: A New Lens on Retrieval Augmented Generation Systems
Considers RAG performance through the lens of sufficient context and explores methods for improving Large Language Model (LLM) factuality (e.g., selective generation)
Hailey Joren, Jianyi Zhang, Chun-Sung Ferng, Da-Cheng Juan, Ankur Taly, Cyrus Rashtchian
Under Review

Classification with Conceptual Safeguards
Introduces conceptual safeguards, a method for selective classification in concept bottleneck models that can reliably promote safety and interpretability in automation tasks using uncertainty propagation
Hailey Joren, Charles T. Marx, Berk Ustun
ICLR 2024

Participatory Personalization in Classification
Introduces a family of prediction models that preserves privacy while letting individuals opt into personalization at inference time
Hailey Joren, Chirag Nagpal, Katherine Heller, Berk Ustun
NeurIPS 2023
Spotlight Recognition

DYffusion: A Dynamics-informed Diffusion Model for Spatiotemporal Forecasting
Proposes a diffusion model tailored for probabilistic spatiotemporal forecasting with specially-designed temporal forward and reverse processes
Salva Rühling Cachay, Bo Zhao, Hailey Joren, Rose Yu
NeurIPS 2023

Learning Document Graphs with Attention for Image Manipulation Detection
Introduces attention mechanisms for graph anomaly detection using pre-trained variational autoencoder representations with graph neural networks (GNNs)
Hailey Joren, Otkrist Gupta, Dan Raviv
ICPRAI International Conference on Pattern Recognition and Artificial Intelligence 2022

MRZ code extraction from visa and passport documents using convolutional neural networks
Proposes an image segmentation model for MRZ code extraction
Yichuan Liu, Hailey Joren, Otkrist Gupta, Dan Raviv
International Journal on Document Analysis and Recognition (IJDAR) 2022

Printing and Scanning Investigation for Image Counter Forensics
Explores adversarial attacks using printing and scanning hardware for document forgery detection
Hailey Joren, Otkrist Gupta, Dan Raviv
EURASIP Journal on Image and Video Processing 2021

Participatory Systems for Personalized Prediction
Introduces prediction methods that allow for partial personalization using model multiplicity
Hailey Joren, Chirag Nagpal, Katherine Heller, Berk Ustun
A Participatory Approach to AI for Mental Health Workshop at NeurIPS 2022
Selected for Oral Presentation

Probabilistic Bias Mitigation in Word Embeddings
Proposes a method for reducing bias in embeddings for downstream natural language processing (NLP) tasks
Hailey Joren, David Alvarez-Melis
Workshop on Human-Centric Machine Learning at NeurIPS 2019

Academic Service and Teaching

Graduate Women in Computing (GradWIC) Mentorship Director

I ran the GradWIC mentorship program. With over 150 participants, I match incoming students from underrepresented backgrounds with experienced graduate student mentors and supervise mentorship across all mentor pairings.

Teaching

I volunteered as a lecturer and teaching assistant for AddisCoder, a summer program that teaches computer science to high school students in Ethiopia.

I served as a teaching assistant for CSE258: Recommendation System and Web Mining, taught by Julian McAuley.

During college, I served as a Harvard Patel Fellow, the recipient of a fellowship that allowed me to work with teaching staff to support students of diverse backgrounds and work individually with students in an introductory computer science course.

Reviewing

ICLR 2024
NeurIPS 2023
NeurIPS 2022