Zan Armstrong

Information is Beautiful - Silver

Why are So Many Babies Born Around 8am

Scientific American

It is common to aggregate data over time when striving for simplicity, but sometimes the story is revealed in the details: minutes, hours, days, and weeks.

Ars Electronica: Understanding AI

Activation Atlas

Distill.pub

The Activation Atlas helps us see through the eyes of a model, not just how it responds to a specific input but to provide a map of what is "seeable" at each layer. This work was led by Shan Carter; I contributed to the interactives and .

SF Moma, Cooper Hewitt, Cube

Metagenomics

with Stamen Design, for UC Berkeley

Originally created for researchers at UC Berkeley to study the genetics of micro-ecosystems like the gut of an infant, the inside of a nuclear reactor, or between a dolphin's teeth, this visualization was also embraced by the art & design community.

My favorite part were the bold colors. They met a scientific need: despite being close numerically, the values 0 and 1 were shown in black and bright colors, respectively, due to the scientific context that "none" and "at least 1" were conceptual opposites.

Make "the Important" Visible

Nightingale Magazine

Published in the "Guidelines" issue of the Data Visualization's Nightingale Magazine, this article shows readers how to make more effective data visualizations by ensuring that "whatever is most important about their data" is actually made literally visible in some way. It's not as obvious as it seems, but when applied it is extremely powerful.

Everything is Seasonal

OpenVisConf

In this talk you'll learn why your commute is worse in November than June, why so many babies are born around 8:30am, and why it's almost always a terrible idea to look at monthly data.

Stop Aggregating Away the Signal

Stack Overflow's blog "The Overflow"

Published on Stack Overflow's blog, this article shows how "by aggregating our data in an effort to simplify it, we lose the signal and the context we need to make sense of what we’re seeing."

Three Simple, Flexible "Tools"

Outlier, Data Visualization Society

This talk introduces the three simple, flexible "tools" of data visualization: making color meaningful, using small multiples, and making order matter. Futhermore, it shows *why* these tools are so powerful to reveal information that might have been visible, but not noticeable or attention-grabbing. Most importantly, it demonstrates how practioners can use these principles to bring their own domain expertise or understanding of the question into their visualizations, and to make their implicit assumptions explicit.

Research: Visualizing Mix

IEEE InfoVis

Simpson's Paradox is an extreme version of a notoriously confusing effect in which aggregate numbers seem to contradict more detailed statistics. A "comet chart" is a special type of scatterplot to reveal "changes in mix". In addition to showing the metric of interest on the y-axis, it also shows the size of each subgroup on the x-axis. It's immediately obvious if comets are all "streaking together", going all different directions, or if there is some correlation between where the comets are spacially and how they are changing.

The Shapes of Emotions

with Stamen Design, for the Dalai Lama

Telling the story of using math to create shapes that convey feelings.

Searching for New Malaria Drugs

Google Applied Science: Malaria

Over the course of three years, I worked closely with scientists and ML experts as they built models to create new methods in the effort to find new anti-malarial drugs.

In the scientific context, authorship is reserved for contributions to the scientific results. So, my inclusion as an contributing author indicates the critical role visualization played in this work.

Analying Time Series

Observable: illustrating techniques

At Observable, I led this project to illustrate how we can literally change how we look at our data to more effectively analyze data heavily influenced by hour-of-day, day-of-week, and season-of-year patterns. This work was done in collaboration with Ian Johnson and Mike Freeman. Through a series of 6 "stories" we showcase techniques that can be applied to any dataset with values measured each hour or day.

SARS-CoV-2 in Wastewater

Stanford University

When infected with a disease, we poop fragments of the viral RNA. Sampling wastewater and isolating/sequencing the extracted genes therefore provides unique insight into community-level transmission of infectious diseases, including Covid-19. Because "everybody contributes" their poop to the sewage system, tracking viral fragments in wastewater provides data that doesn't rely on individual people going to doctors to get official tests.

This LA Times article demonstrates the impact that this data has already had on the decisions of public health officials and of doctors making decisions about which drugs they should use to treat patients based on which variant is predominant.

Changing How Scientists Create Viz

Google: Developing & Teaching Best Practices

My work is not just about specific charts, but also about identifying new best practices and changing how researchers think about using viz more effectively in their own work.

Gene Expression in the Brain

Yale University

The design emphasized the importance of the two features that made this dataset special: quantitative accuracy and gene expression assayed by cell type.

This visualization played a role in the analysis leading to A multiregional proteomic survey of the postnatal human brain published in Nature Neurosience

Contributing to Top Tier Research

Google Accelerated Science: Materials

My first attempt was a beautiful, interactive 3D surface. Unfortunately, it was useless for analyzing the data. Instead, these 48 line charts show 10x more data and focussed attention on the most important attributes of the data. Most importantly, the tool informed key research decisions and led to discovering an unexpected scientific phenomenon.

My contributions led to being an author in scientific research published in published in Proceedings of the National Academy of Sciences (PNAS), the world's second most cited scientific journal.

Informing ML feature space

Google: Viz + ML

A common problem in ML is understanding the interconnected relationships between 3 to 20 potential features in a dataset to identify what combinations of features most affect a "score" variable. My key data visualization insight was identifying that if we discretized the continuous score function into "bad, meh, good, better, and best" categories we could much more effectively use color and small multiples to scan the space of pairwise interactions. This was especially important in this study where the ML algorithms informed growing conditions for algae, so measuring the effectiveness of a particular set of parameters was costly and time-consuming because it required actually growing the algae in the lab.

Viz for Scientific Communication

Google: Advising

While my primary role on the Applied Sciences team was creating visualizations for analysis & discovery, I also advised researchers across the team on how to more effectively communicate their results.

The "curse of knowledge" bias is especially challenging in transitioning from analysis to communication: once you know what's important about the data in the chart, it's hard to imagine that someone else looking at the same chart wouldn't see the same thing.

Covid Response

Google: Data Viz Hub

In March or 2020, I founded and led the Data Viz Hub: a group of Googlers with data viz skills lending their expertise to projects across Google related to Covid-19. For example, our group contributed to the Google Search and News teams' visualizations of covid statistics.

Additionally, Adam Pearce and I collaborated to create visualizations and thought questions to contextualize the data in the Google's Community Mobility Reports.

Statistics Textbook

Published by Wiley

I developed practice exercises for the data visualization chapter of this college-level textbook: Statistics: Unlocking the Power of Data.

A Jittery Line

Because it's delightful

These jittering lines still make me smile, even long after I first explored animation as a student at the School for Poetic Computation.

Voronoi Diagrams Explorable

Drawing with math!

Using voronoi polygons to capture click events in Weather Lines sparked my curiosity. The result was an "explorable" that reveals the relationship among points, the bisecting lines that form triangles, circles, and the resulting polygons.

Which is Bigger?

Created because Africa is really big

Everyone who's watched West Wing knows that Greenland is not as big as it seems in mercator projections. But, do you have a good intuition for the relative sizes of Saudi Arabia vs. Alaska or Europe vs. Antarctica?

Perceptual Uniformity vs. Distance

What is the trade-off?

This set of charts explores the relationship between perceptual uniformity and perceptual distance for 11 common colormaps, and it illustrates an alternative way to achieve uniformity by "straightening" the colormap.

D3 Number Format "Tiny Tool"

My 15 minutes of fame

Long ago I shared two rainbow colored "tiny tools" that I'd created to make it easier to choose formatting strings, and they ended up being surprisingly helpful to many people.

Weather Circles (and Lines)

Mark Twain was wrong.

I lost a bet. It is colder in January than June in San Francisco, every single hour of the day.

High Contrast Cards for Babies

Training neural nets for edge detection

Young babies love looking at high contrast cards, and the designs make viz-loving new parents smile.

PS - for V2, would like to print on sturdier materials. Contact me if you know how to order custom prints on materials a baby can put in their mouth.

Guest Speaker

Teaching and Workshops

Interviews