First day of school! Today I'm starting as a professor at Harvard. My research will center on visualization, human/AI interaction, and how we can use computers to generate insight.
Tiny neural networks have a surprisingly rich inner life, and may hold clues to how their larger cousins work. This image: evolution of feature vectors during learning. Full story: (in collaboration with the great interpretability team at
@AnthopicAI
)
PAIR's new "What-If Tool" lets you probe a machine learning model, no extra coding required! Visualize inference, find counterfactuals, see feature importance, try algorithmic fairness criteria. Now in TensorBoard!
@bengiswex
@mahimapushkarna
Jimbo Wilson
Visualize transformer attention!
AttentionViz, created by Catherine Yeh and expanded by Yida Chen, helps you explore transformer self-attention by visualizing query and key vectors in a joint embedding.
Paper:
Website:
I’m excited to announce that starting in fall 2021 I’ll be a professor of computer science at Harvard (and affiliated with the business school). And yes,
@viegasf
will join me in spring 2022! We’ll also continue to work part-time at Google and on PAIR.
Thinking about grad school next year? Interested in visualization, human/AI interaction, or computational art? Consider Harvard!
@viegasf
and I are building up our lab and will be looking for students. Our DMs are open.
When
@viegasf
and I designed our wind map, we scaled the brightness by the top windspeed in the data. We also only plot wind over land.
So when a hurricane is heading to landfall, the map goes ominously, unreadably dark. Technical bug, emotional feature.
I've been seeing unusual "tornado plot" diagrams floating around, and wrote a tiny interactive tool to help myself understand them.
Draw your own x-y graph, see what it turns into with a tornado plot! (Spoiler: it's probably loopy.)
Looking for a postdoc position in data visualization, human-centered machine learning, or both? Consider joining our lab at Harvard! Here’s the official announcement:
Thinking about grad school next year? Interested in visualization, machine learning interpretability, or human/AI interaction? Consider Harvard.
@viegasf
and I are continuing to build our lab!
UMAP is a real advance in visualizing high dimensional data. But how should we read it? And when should we use it? Take a look at this interactive essay by
@_coenen
and
@adamrpearce
!
Machine learning on the web takes the next step... Awesome improvements to deeplearn.js - now known as TensorFlow.js. Can't wait to see what people do with this!
Why people describe warm colors more efficiently than cool ones—across 110 languages! Nice research on color terms, with elegant visualizations and a new use for a classic ML data set.
Paper:
Summary: (h/t
@timhwang
&
@emollick
)
Wonderful essay on science and art! Jane Richardson describes how woodcuts, vases, and origami influenced her research. Now I know where those "ribbon drawings" of proteins come from.
I recreated Byrne's 1847 edition of Euclid's Elements including interactive diagrams, cross references, and a new poster of all the original illustrations: Plus an in-depth blog post on how it was made:
Can we predict earthquake aftershock locations with AI? Will machine learning and science go together like chocolate and peanut butter?
In Nature today:
Or see:
(collaboration with Phoebe DeVries,
@brendanjmeade
, and
@viegasf
)
Highly recommend Code as Creative Medium by
@golan
and
@tegabrain
. It’s a genuinely important book for anyone interested in teaching or learning about algorithms and art. A wonderful mix of theory, practice, and inspiration!
Designing with AI requires a big change in mindset—in my opinion, just as big as when graphical user interfaces were introduced.
I'm delighted to announce the People + AI Guidebook, which lays out recommendations for designing human-first AI products.
The blog at explores the shape of these representations, with math and visualization. A mysterious “squared distance” effect they found turns out to be surprisingly natural!
Some context: Recently
@johnhewtt
and
@chrmanning
discovered certain neural networks (e.g., BERT) build internal geometric representations of syntax trees. It’s beautiful: the full tree is actually embedded in a high-dimensional vector space.
This generative art walkthrough is so good!
The visuals are a treat, the explanations clear and generous.
(Art and essay by
@tylerxhobbs
. h/t
@soficrespo91
)
What does a neural net see when it looks at an Escher print?
Explore illusions in an automatic depth perception system:
From
@EllenJiang2
,
@emilyrreif
and
@_beenkim
Look at all the ways people draw water in maps! This is just one piece from a beautiful set of tweets spanning design, history, coding, and math—I recommend the whole thread, and the set of notebooks at
What does it really mean to design for human/AI collaboration? How can doctors and ML work together effectively? If you're at CHI, go to this talk for some answers! (And if you're not, the paper's online.)
How can human-centered tools make AI more useful to doctors? I'll be presenting on our
#chi2019
paper tomorrow, which won a best paper honorable mention!
Human-Centered Tools for Coping with Imperfect Algorithms during Medical Decision-making
Monday 11am
Great thread. A good visualization doesn't just transmit data, it also helps people mentally compute with that data. Designers should consider both aspects.
This paper sheds light on how different visual encodings affect one common computation people do, namely averaging.
IN PRESS: Perceptual proxies for extracting averages in data visualizations
by Lei Yuan,
@sharoz
, &
@SteveFranconeri
to appear in Psychonomic Bulletin and Review
This work asks how we compare set averages in graphs. 1/
#dataviz
#visionscience
#ieeevis
Redesign thread: This chart in today's
@nytmag
is attractive and interesting, but the design gives outer rings way more visual weight than inner ones--and really, outer vs. inner shouldn't matter. I started wondering how much difference this actually makes.
After reading this article, I wonder... if the 1918 flu had left a bigger cultural footprint, would the world have been more attuned to the risks of a pandemic?
Bach to future! Compose your own music, with the help of the AI on today's home page. Great collaboration with the Magenta, Doodle and TensorFlow.js teams. Congrats to all involved!
I wish we talked about global warming in millidegrees. A mere 1.1 degree increase is easy to ignore. A whopping 1,100 millidegrees might feel more like a crisis.
Toasters have blinking lights, cars have speedometers. Should chatbots have dashboards too?
A speculative essay: The System Model and the User Model: Exploring AI Dashboard Design
Language models have some beautiful spiral plots reflecting positional patterns. And a vision model has heads that arrange images according to brightness and hues. But there’s a lot more to find! What else can you see?
@EllenJiang2
@emilyrreif
@_beenkim
And what happens when you apply this idea to other artworks across history?
Read more here:
And try it yourself with this interactive visualization:
I'm teaching with
@OpenProcessing
for the first time, and am completely impressed with how polished and friendly the system is. Every detail is on point. Last class a student spontaneously said, "OpenProcessing is just so great." I agree!
A thread on using
#midjourney
to create illustrations for a talk.
Very different from art / exploration / whimsy, and an interesting example of human-AI interaction!
Every cover of The New Yorker, from 1925 to 2018.
The horizontal line/artifact comes from the way that they re-run their first cover (and more recently, a variant of it) each year on their anniversary.
Much better up close:
A shiny new name grapher!
Full description here:
For visualization aficionados: compare the line vs. stacked graph views. Such a contrast, and useful in different ways.
I also made a bar chart, for comparison. It's definitely more precise. But as I made it, I realized that what I really wanted to know was, what age do people think it's OK to have a phone? So I made the cumulative chart on the right.