Kaidi Cao

I am a final-year Computer Science Ph.D. candidate at Stanford University, advised by Prof. Jure Leskovec. I received B.E. from Tsinghua University.

My research interests lie generally in the area of machine learning, with focus on efficient, robust learning algorithms. I am also interested in retrieval-augmented LLMs and multi-modal alignment.

Email / LinkedIn / Google Scholar / Github

Learning Large Graph Property Prediction via Graph Segment Training
Kaidi Cao, Phitchaya Mangpo Phothilimthana, Sami Abu-El-Haija, Dustin Zelle, Yanqi Zhou, Charith Mendis, Jure Leskovec, Bryan Perozzi
Neural Information Processing Systems (NeurIPS), 2023
[pdf] [code]

We propose Graph Segment Training (GST), which is able to train on large graphs with constant (GPU) memory footprint.

TpuGraphs: A Performance Prediction Dataset on Large Tensor Computational Graphs
Phitchaya Mangpo Phothilimthana, Sami Abu-El-Haija, Kaidi Cao, Bahare Fatemi, Charith Mendis, Bryan Perozzi
Neural Information Processing Systems (NeurIPS), 2023
[pdf]

We present a dataset for TPU-based performance prediction, offering new scalability challenges.

AutoTransfer: AutoML with Knowledge Transfer - An Application to Graph Neural Networks
Kaidi Cao, Jiaxuan You, Jiaju Liu, Jure Leskovec
International Conference on Learning Representations (ICLR), 2023
[pdf]

We propose AutoTransfer, an AutoML solution that improves search efficiency by transferring the known architectural design knowledge to the novel task of interest.

Annotation of Spatially Resolved Single-cell Data with STELLAR
Maria Brbić*, Kaidi Cao*, John W. Hickey*, Yuqi Tan, Michael P. Snyder, Garry P. Nolan & Jure Leskovec
Nature Methods, 2022
[pdf] [code]

We present STELLAR, a geometric deep learning method for cell-type discovery and identification in spatially resolved single-cell datasets.

Relational Multi-Task Learning: Modeling Relations between Data and Tasks
Kaidi Cao*, Jiaxuan You*, Jure Leskovec
International Conference on Learning Representations (ICLR), 2022
[pdf] [code]

We propose MetaLink to solve a variety of multi-task learning settings, by constructing a knowledge graph over data points and tasks.

Open-World Semi-Supervised Learning
Kaidi Cao*, Maria Brbić*, Jure Leskovec
International Conference on Learning Representations (ICLR), 2022
[pdf] [code]

We propose a pipeline that recognizes previously seen classes and discovers novel, never-before-seen classes at the same time.

Heteroskedastic and Imbalanced Deep Learning with Adaptive Regularization
Kaidi Cao, Yining Chen, Junwei Lu, Nikos Arechiga, Adrien Gaidon, Tengyu Ma
International Conference on Learning Representations (ICLR), 2021
[pdf] [code]

We propose a data-dependent regularization technique for heteroskedastic and imbalanced datasets.

Concept Learners for Few-Shot Learning
Kaidi Cao*, Maria Brbić*, Jure Leskovec
International Conference on Learning Representations (ICLR), 2021
[pdf] [code]

COMET learns generalizable representations along human-understandable concept dimensions.

Coresets for Robust Training of Neural Networks against Noisy Labels
Baharan Mirzasoleiman, Kaidi Cao, Jure Leskovec
Neural Information Processing Systems (NeurIPS), 2020
[pdf] [code]

We propose a theoretically-principled method to create sets of clean data to train a model with noisy labels.

Few-Shot Video Classification via Temporal Alignment
Kaidi Cao, Jingwei Ji*, Zhangjie Cao*, Chien-Yi Chang, Juan Carlos Niebles
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2020
[pdf] [split]

We propose a video few-shot learning framework that explicitly leverages the temporal ordering information in video data through temporal alignment.

Learning Imbalanced Datasets with Label-Distribution-Aware Margin Loss
Kaidi Cao, Colin Wei, Adrien Gaidon, Nikos Arechiga, Tengyu Ma
Neural Information Processing Systems (NeurIPS), 2019
Oral presentation at the Bay Area Machine Learning Symposium
[pdf] [code]

We design two novel methods to improve imbalanced training.

CariGANs: Unpaired Photo-to-Caricature Translation
Kaidi Cao, Jing Liao, Lu Yuan
ACM Transactions on Graphics, (Proc. of Siggraph Asia), 2018
[pdf] [project page]

We present the first deep learning-based approach to automatically generate the facial caricature for a given portrait photo.

Press Coverage:

Pose-Robust Face Recognition via Deep Residual Equivariant Mapping
Kaidi Cao*, Yu Rong*, Cheng Li, Xiaoou Tang, Chen Change Loy
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018
[pdf][project page][code]

We presented a Deep Residual EquivAriant Mapping (DREAM) block to improve the performance of face recognition on profile faces.

Merge or Not? Learning to Group Faces via Imitation Learning
Yue He*, Kaidi Cao*, Cheng Li, Chen Change Loy
AAAI Conference on Artificial Intelligence (AAAI), 2018
[pdf][code]

We proposed a novel face grouping framework that makes sequential merging decision based on short- and long-term rewards via inverse reinforcement learning.