Here is a list of sample papers I have been written over the years.
One of the major themes I have worked is around better understanding the generalization ability of modern neural networks.
In particular, this also touches on the role of optimization algorithms, and statistical modeling of modern learning paradigms.
Noise Stability Optimization for Finding Flat Minima: A Hessian-based Regularization Approach. H. R. Zhang, D. Li, and H. Ju. Transactions on Machine Learning Research (TMLR), 2024
On the Generalization Effects of Linear Transformations in Data Augmentation. S. Wu*, H. R. Zhang*, G. Valiant, and C. Ré. International Conference on Machine Learning (ICML), 2020
Algorithmic Regularization in Over-parameterized Matrix Sensing and Neural Networks with Quadratic Activations. Y. Li*, T. Ma*, and H. Zhang*. Annual Conference on Learning Theory (COLT), 2018
Here is my talk slide about this line of work.
Another major theme of my work has been on developing efficient algorithms for foundation models.
This has involved studying the multitasking capability of large neural networks, and developing efficient fine-tuning algorithms for adapting these models.
Scalable Fine-tuning From Multiple Data Sources: A First-order Approximation Approach. D. Li, Z. Zhang, L. Wang, and H. R. Zhang. Findings of Empirical Methods in Natural Language Processing (EMNLP), 2024
Identification of Negative Transfers in Multitask Learning Using Surrogate Models. D. Li, H. L. Nguyen, and H. R. Zhang. Transactions on Machine Learning Research (TMLR), 2023. Featured Certification
Robust Fine-Tuning of Deep Neural Networks with Hessian-based Generalization Guarantees. H. Ju, D. Li, and H. R. Zhang. International Conference on Machine Learning (ICML), 2022
Understanding and Improving Information Transfer in Multi-Task Learning. S. Wu*, H. R. Zhang*, and C. Ré. International Conference on Learning Representations (ICLR), 2020
Here is a presentation slide that I have used to describe this line of work recently.
I have also been working on graph neural networks, including their sample complexities. Recently we collected a traffic accident dataset where we found that using GNNs we can give accurate predictions of accident occurrences.
Generalization in Graph Neural Networks: Improved PAC-Bayesian Bounds on Graph Diffusion. H. Ju, D. Li, A. Sharma, and H. R. Zhang. Artificial Intelligence and Statistics (AISTATS), 2023
Graph Neural Networks for Road Safety Modeling: Datasets and Evaluations for Accident Analysis. A. Nippani, D. Li, H. Ju, H. N. Koutsopoulos, and H. R. Zhang. Neural Information Processing Systems (NeurIPS Datasets Track), 2023
Here are a list of manuscripts I have been currently working on. They are either under review or under revision at a conference or a journal.
Precise High-Dimensional Asymptotics for Quantifying Heterogeneous Transfers. F. Yang, H. R. Zhang, S. Wu, C. Ré, and W. Su
I coordinate a mailing list about machine learning seminar within Northeastern and we are always looking for interested folks to speak. Feel free to contact us if you are interested in presenting your work at Northeastern! We are located at 177 Huntington Ave, Boston.