Jun Yuan (袁珺)

CV  /  Google Scholar  /  LinkedIn

profile photo
News

I'm going to attend IEEE VIS, see you all in Florida!

[June 2024] I'm giving a guest lecture at University of California, Berkeley, class DS100 Data, for “Visualization II”.

[Oct. 2023] I'm presenting our paper SuRE and SUBPLEX in VIS'23.

[Mar. 2023] I moved to CA and started working at Apple .

[Feb. 2023] I passed my PhD dissertation defense!

Research
Visual Exploration of Machine Learning Model Behavior with Hierarchical Surrogate Rule Sets
Jun Yuan​, Brian Barr, Kyle Overton, Enrico Bertini.
IEEE Transactions on Visualization and Computer Graphics (TVCG)
arXiv /  video /  github /  evaluation

In this paper, we contribute SURE, a visual analytics system that integrates an algorithmic and interactive solution to generate and visualize surrogate rules to help people understand model behaviors. Our validation studies with this system leads to many interesting findings, including a task taxonomy for rule analysis, that can be used for visualization design and future research.

iSEA: An Interactive Pipeline of Semantic Error Analysis for NLP Models
Jun Yuan​, Jesse Vig, Nazneen Rajani.
ACM IUI, 2022  
arXiv /  paper /  demo /  github

iSEA is an interactive pipeline for Semantic Error Analysis in NLP Models, which automatically discovers semantically-grounded subpopulations with high error rates in the context of a human-in-the-loop interactive system.

SUBPLEX: A Visual Analytics Approach to Understand Local Model Explanations at the Subpopulation Level
Jun Yuan​, Gromit Yeuk-Yin Chan, Kyle Overton, Brian Brian, Kim Rees, Luis Gustavo Nonato, Enrico Bertini, Cláudio T. Silva.
IEEE Computer Graphics & Applications
(Special Issue on Human-Centered Visualization Approaches to AI Explainability)
, 2022  
demo video  /  paper

SUBPLEX is an interactive visualization widget embedding in Jupyter notebook that enables users to explore and validate local explanations of a model at the sub-population level.

Context Sight: Model Understanding and Debugging via Interpretable Context
Jun Yuan​, Enrico Bertini.
Workshop on Human-In-the-Loop Data Analytics (HILDA), ACM SIGMOD, 2022  
paper /  talk /  slides

In this work, we first identify two main components of context to assist model diagnosis through literature reviews. We then present Context Sight, a visual analytics system that integrates customized context generation and multiple-level context summarization.

An Exploration and Validation of Visual Factors in Understanding Classification Rule Sets
Jun Yuan​, Oded Nov, Enrico Bertini. IEEE VIS, 2021
arXiv /  paper

In this paper, we explore the idea of designing alternative representations of rules, focusing on a number of visualfactors we believe have a positive impact on rule readability andunderstanding. We then presents a user study exploring their impact.

AdViCE: Aggregated Visual Counterfactual Explanations for Machine Learning Model Validation
Oscar Gomez, Steffen Holter, Jun Yuan​, Enrico Bertini. IEEE VIS, 2021
arXiv /  paper

AdViCE is a visual analytics tool that aims to guide users in black-box model debugging and validation through counterfactuals.

Convolution Can Incur Foveation Effects
Jun Yuan​, Bilal Alsallakh, Narine Kokhlikyan, Vivek Miglani, Orion Reblitz-Richardson.
Rethinking ML Papers - ICLR 2021 workshop, 2021  
website  /  video

Mind the Pad -- CNNs Can Develop Blind Spots
Bilal Alsallakh, Narine Kokhlikyan, Vivek Miglani, Jun Yuan​, Orion Reblitz-Richardson.
ICLR, 2021   (Spotlight)
arXiv  /  paper
mTSeer: Interactive Visual Exploration of Models on Multivariate Time-series Forecast
Ke Xu, Jun Yuan​, Yifang Wang, Claudio Silva, Enrico Bertini.
SIGCHI, 2021  
paper  /  video

mTSeer is an interactive system for the exploration, explanation, and evaluation of multivariate time-series forecasting models.

ViCE: Visual Counterfactual Explanations for Machine Learning Models
Steffen Holter, Oscar Gomez, Jun Yuan​, Enrico Bertini.
ACM IUI, 2020  
paper  /  code

ViCE is an interactive visual analytics system integrating counterfactual explanation generation and exploration to contextualize and evaluate model decisions.

ECGLens: Interactive ECG Classification and Exploration
Jun Yuan​, Siyao Fang, Xiang Huang, Nan Cao.
IEEE VIS, poster, 2017  
paper  /  video

ECGLens is an interactive visual analytics system assisting users to inspect anomalous ECG signals (heartbeat, time-series).

Talks

Apr. 2022, invited lecture at The College of William & Mary, class CSCI780 Data Visualization, "Visualization for Machine Learning Explanations".

Oct. 2020, presentation at Doctoral Colloquium of IEEE VIS 2020, "Interpreting Black-box Machine Learning Models By Visually Exploring High-Fidelity Surrogate Rules".

Service
Program Committee VIS Short Papers (22, 23, 24), PacificVis Vis-Meets-AI Workshop (24)
Reviewer TVCG, CG&A, SIGCHI (19, 22, 23, 24), VIS (19, 20, 22, 23, 24), CSCW (20, 21), ICML (20), ChinaVis (22)
Student Volunteer VIS (20, 21, 22)
Mentor (for Research) NYU Undergradate Research Program (19,20,21)
Mentor (for K12 STEM Education) ARISE (18, 20), 1000 Girls, 1000 Futures (18)
Teaching (NYU)
22'Fall, 19'Fall Graduate Student Instructor CS-GY 6313 Information Visualization
22'Spring Graduate Student Instructor CS-GY 9223 Visualization for Machine Learning
20'Spring, 19'Spring Graduate Student Instructor CS-GY 6323 Visual Analytics
Teaching (Fudan)
17'Spring Teaching Assistant Network Virtual Environment and Computer Application
16'Fall Teaching Assistant Programming





Website adapted from Jon Barron