About Me   

- I like board games, sim racing, and music.
- I worked at Baidu, received my PhD degree at Stony Brook University.
- Now I work in Google Core ML.
- Research of interests: deep learning, computer vision, natural language processing.

lehou [at] google.com
google scholar page

News

"Scaling Instruction-Finetuned Language Models" is out on arXiv.

1. Instruction finetuning language models on a massive number of tasks (1,800+).
2. We have FLAN-PaLM (8 to 540 billion parameters), FLAN-T5, FLAN-UL2, etc.
3. Yes, we got state-of-the-art results!
4. We got significantly better zero-shot and few-shot prompting performance for every model!
5. FLAN-PaLM can just explain its predictions (zero-shot chain-of-thought).

Datasets and Code

- Code for Token Dropping for Efficient BERT Pretraining. ACL 2022.
- SBU shadow dataset for Large Scale Shadow Annotation And Detection Using Lazy Annotation and Stacked CNNs. PAMI 2019.
- Code and data for Robust Histopathology Image Analysis: to Label or to Synthesize? Oral in CVPR. 2019.
- Individual lymphocyte classification dataset, and code and data and a Pytorch implementation (Thanks to Mihir Sahasrabudhe) for Sparse Autoencoder for Unsupervised Nucleus Detection and Representation in Histopathology Images. Pattern Recognition. 2018.
- Code for Spatial Organization and Molecular Correlation of Tumor-Infiltrating Lymphocytes Using Deep Learning on Pathology Images. Cell Reports. 2018.
- Code for Squared Earth Mover's Distance Loss for Training Deep Neural Networks on Ordered-Classes. NeurIPS workshop on Learning on Distributions, Functions, Graphs and Groups. 2017.
- Code for ConvNets with Smooth Adaptive Activation Functions for Regression. AISTATS. 2017.
- Nucleus classification dataset for Center-Focusing Multi-task CNN with Injected Features for Classification of Glioma Nuclear Images. WACV. 2017.