PeopleofSuperintelligence

Researcher

Xuxi Chen

Xuxi Chen is a machine-learning researcher whose work spans positive-unlabeled learning, sparse networks, multilingual translation and language models. He studied statistics at USTC while working with VITA in 2019, then coauthored research on self-training and efficient sparse networks. His later work ranges from Orca 2's reasoning methods for small language models to Lingual-SMoE's multilingual routing and a study of alignment behavior. After completing his VITA PhD in Summer 2025, he is a quantitative researcher at XTX Markets as of October 2026.

Badges

Timeline

  1. 2019

    Statistics at USTC

    During Summer 2019, he works with VITA as an undergraduate in statistics at USTC and plans to join the group for a PhD.

  2. 2020

    Self-PU training

    In 2020, he coauthors Self-PU, a positive-unlabeled learning framework that combines self-training with self-paced training to identify confident positive and negative examples.

  3. 2021

    Efficient lottery-ticket finding

    In 2021, he coauthors Efficient Lottery Ticket Finding: Less Data is More, which uses a Pruning-Aware Critical set of training data to find winning subnetworks with fewer training iterations. In Fall 2021, he begins PhD study with VITA.

  4. 2022

    Sparse tickets in low-data settings

    In 2022, he coauthors Sparse Winning Tickets are Data-Efficient Image Recognizers. In experiments with 50 to 100 examples per class, the paper finds that sparse winning tickets outperform the original dense networks.

  5. 2022

    Structurally sparse lottery tickets

    Also in 2022, he coauthors Coarsening the Granularity: Towards Structurally Sparse Lottery Tickets. The method refills important channels and regroups nonzero elements to create channel- and group-wise sparse networks with inference speedups.

  6. 2023

    Orca 2 reasoning research

    In 2023, he coauthors Microsoft's Orca 2, which teaches small language models to choose different solution strategies for different tasks. Microsoft releases Orca 2 as open source to support further research.

  7. 2024

    Lingual-SMoE for translation

    In 2024, he coauthors Lingual-SMoE for multilingual machine translation. It uses language representations to group experts and adapts the number of active experts for each target language to its translation difficulty.

  8. As of October 2026

    Quantitative researcher at XTX Markets

    In 2025, he coauthors Extracting and Understanding the Superficial Knowledge in Alignment, a study of how alignment fine-tuning changes shallow token selection in language models. He completes his VITA PhD in Summer 2025. As of October 2026, he is a quantitative researcher at XTX Markets.

Career

  1. NowXTX MarketsQuantitative researcher
  2. 2019VITAUndergraduate

Education

  1. 2025VITAPhD
  2. From 2019University of Science and Technology of ChinaUndergraduate

From the biography.

  1. Universities keep AI researchers for a median of 6 years, compared with 2 years at frontier labsIncludes University of Texas at Austin. Across 15,811 dated work stays (6,529 people), median role tenure runs from 6 years at universities to 2 years at frontier labs. We compare all five organization types, starting decades from the 1970s to the 2020s, and the organizations that keep people longest.
  2. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes University of Texas at Austin. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.
  3. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes University of Texas at Austin. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.
  4. People shaping AI work at a median of 3 organizations over a career: 3 for university starters and institute and lab starters, versus 2 for company startersIncludes University of Texas at Austin. Across 6,853 people shaping AI with a dated work role, we estimate a median of 3 parent organizations over a career (mean 3.6), comparing role groups, starting sectors and starting decades from the 1950s to 2020–2026.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Xuxi Chen yet. Send us a free-licensed photo.

  1. Team | VITA Group vita-group.space. Checked 21 October 2026.
  2. Self-PU: Self Boosted and Calibrated Positive-Unlabeled Training proceedings.mlr.press. Checked 21 October 2026.
  3. Efficient Lottery Ticket Finding: Less Data is More proceedings.mlr.press. Checked 21 October 2026.
  4. Sparse Winning Tickets are Data-Efficient Image Recognizers proceedings.neurips.cc. Checked 21 October 2026.
  5. Coarsening the Granularity: Towards Structurally Sparse Lottery Tickets proceedings.mlr.press. Checked 21 October 2026.
  6. Orca-2: Teaching Small Language Models How to Reason microsoft.com. Checked 21 October 2026.
  7. Sparse MoE with Language Guided Routing for Multilingual Machine Translation proceedings.iclr.cc. Checked 21 October 2026.
  8. Extracting and Understanding the Superficial Knowledge in Alignment aclanthology.org. Checked 21 October 2026.