Jana Beck

A deep dive into the specific features of Web Workers that aid in making it possible to perform tasks typically reserved for cloud computing jobs, in the browser.

Jana covers how to bring multi-dimensional computing power to the browser using Web Workers. She focuses on one particular dimensionality reduction algorithm—UMAP—used more and more frequently these days to visualize high-dimensional datasets. She deep dives into the specific features of Web Workers that aid in making it possible to perform tasks typically reserved for cloud computing jobs in the browser, albeit on smaller samples from data sets.

These features include transferable objects, which allow for efficient, zero-copy passing of computed results back to the browser from the Worker and OffscreenCanvas, a bleeding edge browser feature that can let us even perform the visual rendering of the UMAP result in a non-blocking Worker.

2019

dotJS

Escaping flatland: a romance of data science in the browser

DinosaurJS

Data science in the browser: DX & UX

Imagine that you’re locked-in, unable to communicate except by blinking a single functional eye. Your only method of communication is to watch (and listen) as a helper reads through a list of all the letters of the alphabet. When you want to select a letter, you blink. 

Can you imagine this? Can we do more than imagine it? On the modern web, we can! Using React and WebGazer.js (an open-source eye-tracking library for the web), we can approximate this communication technique, the method that Jean-Dominique Bauby used to compose his best-selling memoir The Diving Bell and the Butterfly.

2017

React Rally

React-ing htmlFor=empathy

All you need for a simple bar chart is D3.js, the standard library for interactive data visualization on the web. But for a highly complex visualization - and here "visualization" should be interpreted broadly to include all manner of charts, graphs, and dashboards with interactivity provided through scrolling, paging, filtering, focusing, and zooming - managing the state of the visualization becomes a complex task. And the problem multiplies if the data updates in real time.

This talk introduces a Flux-based strategy to separate the state of the visualization into three stores: one for the configuration of the visualization, one for the data that is the subject of visualization, and a third that bridges - and depends on - the first two and provides methods for accessing exactly the state information needed to render the current configuration of the data visualization using the subset of the current dataset that falls within the domain under inspection.