Elamur Research⌕
☰
home / authors / Wendi Zheng

Wendi Zheng

Wendi Zheng works on large generative models built on the transformer architecture, spanning language, images and video. Their papers here, published between 2021 and 2024, ask how a single large network can be trained to produce fluent text in two languages, or to turn a written prompt into a picture or a moving clip. Recurring methods include converting images into sequences of discrete tokens, autoregressive and blank-infilling objectives, and, more recently, diffusion-based generation. The work sits mainly within Vision & language.

The most cited of these papers is GLM-130B (2022), which describes an openly released dense model of 130 billion parameters for English and Chinese. It departs from the decoder-only GPT design by using the GLM backbone, with bidirectional attention over visible context and training by filling in masked spans. The paper documents both successful and failed training choices. Earlier, CogView (2021) treated general-domain text-to-image generation as a two-stage problem. A VQ-VAE-style tokenizer first maps each picture to a grid of codes, and a 4-billion-parameter transformer then models caption tokens followed by image tokens. That model was pretrained on roughly 30 million Chinese caption–image pairs.

The newest work moves from still images to video. CogVideoX (2024) builds a diffusion transformer aimed at a familiar shortcoming of text-to-video systems. Their output clips tend to be brief, show little motion and follow prompts loosely when those prompts describe several events in order.

Zheng's papers here are associated with Tsinghua University, with bylines from 2022 to 2024. Frequent co-authors include Ming Ding and Zhuoyi Yang, each on three papers, along with Wenyi Hong and Da Yin. Other collaborators include Aohan Zeng and Jie Tang.

Wendi Zheng has 3 papers in our library, in 2021–2024, most of them on Vision & language; often with Ming Ding, Zhuoyi Yang, Da Yin.

3 papers in our library · 2021–2024

current affiliationTianjin University of Science and Technology, Tianjin, China; Chinese Academy of Sciences, Beijing, China; South China Sea Institute Of Oceanology, Guangzhou, China; Tianjin Institute of Industrial Biotechnology, Tianjin, China; State Key Laboratory of Tropical Oceanography, Guangzhou

OpenAlex · ORCID · h-index 18 · 78 works · 1 474 citations OpenAlex, 2026-10-14

3 works in our library · 769 citations in our library · 1 more work in our library · ⭐ Paper Star rank #1589

The profile is the library's match of the name; two people with one name are not always told apart.

Elsewhere: Google Scholar · Semantic Scholar · dblp · ACL Anthology

Affiliations

From the bylines of their papers (OpenAlex, 2026-10-14), ORCID and Wikidata; institutions on papers are matched by machine. 12 more institutions on single bylines are not listed.

Topics

Vision & language1

Works by year

2020s

20241 paper20221 paper20211 paper

Papers

The most cited first (OpenAlex and our library).

  1. GLM-130B: An Open Bilingual Pre-trained Model (2022 · 775 citations)
    Few openly available models at the 100B-parameter scale matched GPT-3, and the methods for training them were mostly undisclosed.
  2. CogView: Mastering Text-to-Image Generation via Transformers (2021 · 446 citations)
    CogView addresses general-domain text-to-image generation. Earlier GAN-based systems did reasonably well on narrow datasets such as CUB but struggled with complex scenes like those in MS COCO. The authors frame the task as maximizing a variational bound on the joint likelihood of text and image. Training happens in…
  3. CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (2024 · 163 citations)
    CogVideoX addresses a persistent weakness of text-to-video systems: clips that are short, nearly static, and poorly matched to prompts describing a sequence of events. The authors build a diffusion transformer around four main components:

Also in our library

1 more work of theirs is in our library without a page here yet; it will appear as we publish them.

Co-authors here

12 of 21 co-authors with a page, the most papers together first.

Cited by

Papers of our library cite their works 769 times (of the 140 884 papers whose bibliographies we read); 72 citing papers with a page here →.

This page in Markdown → https://md.research.elamur.ai/free/en/authors/wendi-zheng/