Kezhi Kong

Career

Kong earned a PhD in the Computer Science Department at the University of Maryland, College Park, advised by Tom Goldstein. Kong is now a research scientist at NVIDIA on the foundation model team in Applied Deep Learning Research, which trains the Nemotron family of models.¹

Research

Kong's research covers large language models, graph representation learning and trustworthy machine learning. Earlier first-author papers include "Robust Optimization as Data Augmentation for Large-scale Graphs" (CVPR 2022), GOAT, a global transformer on large-scale graphs (ICML 2023), and OpenTab, on large language models as open-domain table reasoners (ICLR 2024); Kong also co-authored work on the reliability of watermarks for large language models (ICLR 2024).¹

With Dan Su and Ying Lin, Kong is a co-first author of "Nemotron-CC: Transforming Common Crawl into a Refined Long-Horizon Pretraining Dataset" (December 2024), which uses classifier ensembling and synthetic rephrasing to build a large English pretraining dataset, and co-wrote NVIDIA's January 2025 announcement of the dataset.²³⁴

Work and contributions

• Nemotron-CC Open data

  Co-first author of the Nemotron-CC paper and co-author of NVIDIA's blog post announcing the 6.3-trillion-token Common Crawl pretraining dataset.²³⁴

• GOAT: A Global Transformer on Large-scale Graphs Published research

  First author of the ICML 2023 paper on a global transformer for large-scale graphs.¹

Timeline

1. 2021Co-first-authored VQ-GNN, a framework for scaling graph neural networks with vector quantization (NeurIPS 2021).¹
2. 2022Published "Robust Optimization as Data Augmentation for Large-scale Graphs" as first author at CVPR 2022.¹
3. 2023Published GOAT, a global transformer on large-scale graphs, as first author at ICML 2023.¹
4. 2024Published OpenTab, on LLMs as open-domain table reasoners, as first author at ICLR 2024.¹
5. Dec 3, 2024Posted the Nemotron-CC paper as co-first author with Dan Su and Ying Lin.²
6. Jan 9, 2025Co-authored NVIDIA's technical blog post announcing the Nemotron-CC dataset.⁴

Where they work

• NVIDIA 43 people in the wiki

Related people

• Dan SuAlso at NVIDIA
• Tsung-Yi LinAlso at NVIDIA
• Rama AkkirajuAlso at NVIDIA
• Bryan CatanzaroAlso at NVIDIA
• Yejin ChoiAlso at NVIDIA
• Bill DolanLanguage models · Datasets and data
• Pradeep DasigiLanguage models · Datasets and data
• Ruchir PuriLanguage models · Datasets and data
• Jesse DodgeDatasets and data · Language models
• Zach NussbaumLanguage models · Datasets and data