Augmenting Visualization tools with Automated Design & Recommendation
Host
Arvind Satyanarayan
Abstract:
Visualization is a critical tool for data science. Analysts use plots to explore and understand distributions and relationships in their data. Machine learning developers also use diagrams to understand and communicate complex model structures. Yet visualization authoring requires a lot of manual efforts and non-trivial decisions, demanding that the authors have a lot of expertise, discipline, and time in order to effectively visualize and analyze the data.
My research in human-computer interaction focuses on the design of tools that augment visualization authoring with automated design and recommendation. By automating repetitive parts of authoring while preserving user control to guide the automation, people can leverage their domain knowledge and creativity to achieve their goals more effectively with fewer efforts and human errors. In my PhD dissertation, I have developed new formal languages and systems for chart specification and recommendation including the Vega-Lite visualization grammar and the CompassQL query language. On top of on these languages, I have developed and studied graphical interfaces that enable new forms of recommendation-powered visual data exploration including the Voyager visualization browser and Voyager 2, which blends manual and automated chart authoring in a single tool. To help developers inspect deep learning architecture, I also built a tool that combines automatic layout techniques with user interaction to visualize dataflow graphs of TensorFlow models as a part of TensorBoard, TensorFlow’s official dashboard tool. These projects have won awards at premier academic venues, and are used by the Jupyter/Python data science communities and leading tech companies including Apple, AirBnB, Google, Microsoft, Netflix, Twitter, and Uber.
Bio:
Kanit (Ham) Wongsuphasawat is a research scientist at Apple where he works on visualization and interactive systems for data science and machine learning.
Kanit has a PhD in Computer Science from the University of Washington (UW), where he worked with Jeffrey Heer and the Interactive Data Lab on
visualization tools. Kanit also previously worked at a number of leading data-driven technology companies including Google, Tableau Software, Thomson Reuters, and Trifacta.
Visualization is a critical tool for data science. Analysts use plots to explore and understand distributions and relationships in their data. Machine learning developers also use diagrams to understand and communicate complex model structures. Yet visualization authoring requires a lot of manual efforts and non-trivial decisions, demanding that the authors have a lot of expertise, discipline, and time in order to effectively visualize and analyze the data.
My research in human-computer interaction focuses on the design of tools that augment visualization authoring with automated design and recommendation. By automating repetitive parts of authoring while preserving user control to guide the automation, people can leverage their domain knowledge and creativity to achieve their goals more effectively with fewer efforts and human errors. In my PhD dissertation, I have developed new formal languages and systems for chart specification and recommendation including the Vega-Lite visualization grammar and the CompassQL query language. On top of on these languages, I have developed and studied graphical interfaces that enable new forms of recommendation-powered visual data exploration including the Voyager visualization browser and Voyager 2, which blends manual and automated chart authoring in a single tool. To help developers inspect deep learning architecture, I also built a tool that combines automatic layout techniques with user interaction to visualize dataflow graphs of TensorFlow models as a part of TensorBoard, TensorFlow’s official dashboard tool. These projects have won awards at premier academic venues, and are used by the Jupyter/Python data science communities and leading tech companies including Apple, AirBnB, Google, Microsoft, Netflix, Twitter, and Uber.
Bio:
Kanit (Ham) Wongsuphasawat is a research scientist at Apple where he works on visualization and interactive systems for data science and machine learning.
Kanit has a PhD in Computer Science from the University of Washington (UW), where he worked with Jeffrey Heer and the Interactive Data Lab on
visualization tools. Kanit also previously worked at a number of leading data-driven technology companies including Google, Tableau Software, Thomson Reuters, and Trifacta.