S.G.S.G. - Silly Goose's Space and Geometry.

Hello, curious mind. However you land on this page, Big Welcome.

My name is Mingwei. I study Data Visualization, a blend of computer, math and art. I play with visual artifacts (a.k.a visualizations) that help people (including myself) extract insights from data, communicate ideas through figures, and understand complex systems through space, geometry, and structure. I am currently a postdoc scholar at Tufts University with Prof.Remco Chang in the VALT Lab. I was a postdoc at Vanderbilt University with Prof. Matthew Berger. I received my Ph.D. from the University of Arizona, where I studied data visualization in the HDC Lab under the guidance of Prof. Carlos Scheidegger. Check out the reserach projects I've worked on below. Explore the tabs at the top. I hope you find something interesting.

Stay curious, stay silly :P mw

Projects and Publications

Interpreting AI models via visualizations

  • LatentGandr: Visual Exploration of Generative AI Latent Space via Local Embeddings [arXiv]
  • Visualizing Neural Networks with the Grand Tour [html]
  • Comparing DNNs with UMAP Tour [html] [VISxAI Slides] [arXiv]
  • Concept Lens: Visually Analyzing the Consistency of Semantic Manipulation in GANs [arXiv]
  • CAN: Concept-Aligned Neurons for Visual Comparison of Deep Neural Network Models [EuroVis2024]
  • CUPID: Contextual Understanding of Prompt-conditioned Image Distributions [arXiv] [EuroVis2024]
  • Understanding capsule networks [html]

Interactive visualization tools

  • DimBridge: Interactive Explanation of Visual Patterns in Dimensionality Reductions with Predicate Logic [arXiv] [demo] [video demo]
  • UnProjection: Leveraging Inverse-Projections for Visual Analytics of High-Dimensional Data. We learned inverse mapping of dimensionality reduction plots [demo on MNIST] [demo on FashionMNIST] [arXiv]
  • Neuralcubes: Deep representations for visual data exploration [arXiv]

Scientific Applications of Vis

Graph Layout algorithms, Network Visualizations

  • Graph Drawing via Gradient Descent, (GD)^2 [arXiv] [demo]
  • Multicriteria Scalable Graph Drawing via Stochastic Gradient Descent, (SGD)^2 [arXiv] [GitHub]
  • A scalable method for readable tree layouts [arXiv]
  • Visualizing evolving trees [arXiv]

Visual Perceptions, User Studies