Xinlei Chen

Also known as Chen Xinlei, Xin-Lei Chen, Xin-Lei 欣蕾 Chen 陈, Xin-lei Chen

Xinlei Chen is a computer vision researcher whose work includes web-trained convolutional networks, self-supervised image learning and video models. After coauthoring a 2015 paper on web data, he completed a Ph.D. at Carnegie Mellon's Language Technologies Institute in 2018 and became a research scientist at Facebook. In 2021, he coauthored SimSiam and a theoretical study of non-contrastive learning that received an ICML Outstanding Paper Honorable Mention; he later coauthored masked autoencoders in 2022 and V-JEPA in 2024. He moved from Meta FAIR to xAI in June 2025 as a Member of Technical Staff. As of October 2026, he works at OpenAI on its multimodal frontier.

Now at: OpenAI
Works: 285
Badges: Bridge

Publication numbers from OpenAlex, as of 22 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Timeline

2015

Web images for ConvNets

In 2015, he coauthors Webly Supervised Learning of Convolutional Networks with Abhinav Gupta. Their two-step method uses web data and curriculum learning to train convolutional networks.
2018

A PhD at Carnegie Mellon

He completes a Ph.D. in Language and Information Technology at Carnegie Mellon's Language Technologies Institute in 2018, with Abhinav Gupta as his academic advisor. At graduation, he joins Facebook as a research scientist.
2021

The SimSiam method

In 2021, he coauthors Exploring Simple Siamese Representation Learning with Kaiming He. Their SimSiam method learns meaningful visual representations without negative pairs, large batches or momentum encoders.
2021

An ICML honorable mention

In 2021, he coauthors Understanding self-supervised learning dynamics without contrastive pairs with Yuandong Tian and Surya Ganguli. The study asks why non-contrastive methods avoid representation collapse, and the paper receives an ICML Outstanding Paper Honorable Mention.
2022

Masked autoencoders

In 2022, he coauthors Masked Autoencoders Are Scalable Vision Learners with Kaiming He, Saining Xie, Yanghao Li, Piotr Dollár and Ross Girshick. The method masks random image patches and trains a decoder to reconstruct missing pixels for self-supervised computer vision.
2024

V-JEPA learns from video

In 2024, he coauthors Revisiting Feature Prediction for Learning Visual Representations from Video. The paper introduces V-JEPA, a collection of vision models trained on 2 million videos from public datasets through feature prediction.
June 2025

Moving from Meta to xAI

In June 2025, he moves from Meta FAIR to xAI. He works at xAI as a Member of Technical Staff on multimodal understanding and generation.
As of October 2026

Multimodal work at OpenAI

As of October 2026, he works at OpenAI on its multimodal frontier. Before joining OpenAI, he works as a Member of Technical Staff at xAI and a Research Scientist at Meta FAIR, where he works on visual representation pretraining.