It is one thing to be in front of someone and not be seen. It is quite another to not be in front of someone and to never have your absence noticed.
I think we underestimate the role of the second kind of invisibility in perpetuating inequity.
Hi. I'm Charles Isbell, a proud alum of
@GeorgiaTech
('90) and
@MIT
('98). I'm old enough to have been a few firsts, which says more about the world than me. I'm now a Professor doing machine learning and the very proud Dean of
@gtcomputing
.
#BlackInComputing
#BiCRollCall
So anyway I’m the opening keynote for
@NeurIPSConf
#NeurIPS2020
today. I tried to cleverly leverage the virtual format... and so it will either be the best or worst thing I’ve ever done.
And for no reason whatsoever as far as you know, I’m going to tag a bunch of folks:
I am confused by this discussion. Surely we are all coming into this knowing that learning algorithms (not to mention complicated *systems* that involve multiple steps, parameters, loss functions, etc) have inductive biases going beyond the “biases” in data, right?
ML systems are biased when data is biased.
This face upsampling system makes everyone look white because the network was pretrained on FlickFaceHQ, which mainly contains white people pics.
Train the *exact* same system on a dataset from Senegal, and everyone will look African.
We're honoring our former dean
@isbellHFh
by creating an awards program with his namesake—the Charles L. Isbell Jr. Graduate Fellowships.
Once fully endowed, the fellowships will fund grad student cohorts from underrepresented groups annually.
An image of
@BarackObama
getting upsampled into a white guy is floating around because it illustrates racial bias in
#MachineLearning
. Just in case you think it isn't real, it is, I got the code working locally. Here is me, and here is
@AOC
.
For the record, I a morning person (and I don’t drink coffee). In grad school, it was like having a super power: I’d get work done early then go around preventing others from making progress. I may have single-handedly held us back a full decade.
We’d have flying cars by now.
ONE SHOULD NEVER CONFUSE ONE’S GROWING AWARENESS OF SOMETHING WITH AN ACTUAL GROWTH OF THAT THING.
thank you for coming to my all-caps evergreen ted talk.
I took a Programming Languages course in grad school.
I remember a conversation with a TA about a homework problem:
Me: I got it to work but I am suspicious of my solution.
TA: That’s because your approach solves the Halting Problem.
Me: …
TA: …
Me: …so that’s an A?
It was 1993, youngster, and it *was* written from scratch.
And, yes, I still remember the Tuesday when everyone stopped using altavista all at once for Google.
Am I the only person on here who never took a campus tour before going to college?
I’m beginning to think my life trajectory is very different from the other boys.
When I tell students how to give a talk I say something like the below: you are not writing a mystery novel with a reveal at the end.
Tell me what you’re going to tell me. Tell me. Tell me what you told me.
Yesterday I was advising a student on writing a paper. I tell this advice to my students over and over again, so I thought I might as well make a thread about it:
tl;dr: "Your paper does not need a twist on the third act, so be upfront about what you want to say!"
#phdchat
1/
This one does annoy me. It’s not “too simple”, it’s elegant, you rube. YOU MUST EARN YOUR COMPLEXITY!
Also, don’t tell me that it’s obvious and anyone would have this idea while admitting you never have, and it’s never been published before.
ARGH!!!
Here's my conversation with Charles Isbell (
@isbellHFh
) & Michael Littman (
@mlittmancs
). Two for the price of one, and the result is priceless. I hope to host occasional debates on this podcast in the future, so this was a fun test of having two guests on.
As an aside I was in the audience at MIT when Guy Steele gave a big talk introducing a new programming language called Java. He opened his talk with, "Lately, I've been thinking a lot about emacs."
The crowd cheered for maybe a full two minutes.
I think about that sometimes.
The murders that took place in Atlanta have drawn attention to a disturbing national pattern of violence against Asians and Asian Americans. This hits our community hard at
@gtcomputing
, and we stand together with all of you at GT and beyond.
Okay. Here's what I really think.
As a zero comment though let me say that is that fields differ and units differ.
My observations are personal experiences at top 10 computing units. YMMV. IANAL. Bacon is delicious.
Okay, let's go:
But “law and order” often assumes the specific order we’re talking about as if any other choice is chaos. But it isn’t. Mostly the argument is really about *authority* and the right of those who have it to keep it. The actual fear is not of chaos but of a different order.
I just stumbled across my dissertation.
I must say: the first paragraph is very well put together. It's short, but tells you exactly what you must know. I'll maintain my self-esteem by not reading further.
Meanwhile, every PhD student should read:
A year ago I decided to learn Swift during meetings by building a new comic book tracking program. Last week I decided to add a feature I'd wanted for 20 years.
So… I am the kind of person who has 20,484 comics. More to the point, I'm the kind who knows I have 20,484 comics.
As NeurIPS deadline nears, a wise person said:
When looking back at your career, you really don't want “K papers at NeurIPS” to summarize your contribution to the field
You want your ideas to be so useful that most will just take them for granted and not even bother to cite you
If one more student says something to me like “I got a C but I need an A, can you do something? I don’t know what but it can’t hurt to ask.” I may invent time travel and go back in time and just fail everyone on the planet.
Is it next week yet? I need a nap.
I often think about the future.
In our future utopia no one will work... and in our future dystopia no one will work. How does one even tell which one we're in?
What does it feel like to be wrong?
It feels the same way as it feels to be right.
What does it feel like to *realize* you’re wrong?
Oh, that usually feels bad.
So sometimes Georgia Institute of Technology (
@GeorgiaTech
) is referred to in the press as Georgia Tech University.
Today I read an article affiliating me with “Georgia Technological University”.
I’m not so much irritated as amazed by the creativity.
Historically our students complain bitterly about having to take communication classes. Also historically our alum about five years out complain bitterly that they didn't take more communication classes.
So, it's a feature that we have ML algorithms that learn their own features and so (let's say) might minimize designer bias...
...but that's also a bug: it means the mechanism for encoding domain knowledge is finding perfect data.
Is that an optimal, *ahem*, design choice?
I usually frame it differently: it is very common among those of us who enjoy solving puzzles for a living to try to solve a problem when it’s put before us without asking whether it’s the right problem to solve. We tend to absorb the unspoken premises and dive in.
OK. I'm going to throw something out there and a lot of you are not going to want to hear it.
Sometimes decision makers hire people (like physicists and early career data scientists) because they want people who won't ask a lot of "why?" questions and will just follow orders.
Imagine a world where reviewers for AI conferences expect easily runnable code which reproduces the graphs in the paper.
Imagine being able to run the code of arbitrary papers to check understanding, robustness, and to experiment.
The resulting field seems unrecognizable.
@isbellHFh
and I are thrilled that we were both named as ACM Fellows today... a great honor in the CS field. (We nominated each other. Because we're friends and we look out for each other.) Thought our
#omscs
@GTOMSCSCS
peeps might appreciate the news. :-)
@mattyglesias
I think it’s more accurate to say that it would be better off if we scheduled fewer meetings altogether, using Zoom where it makes sense and using other appropriate ways of communicating otherwise.
We are overscheduled, and because of that we schedule more.
From when my boy was 6:
6: Dad, when I grow up will you be old?
- yes….
6: when I'm old will you be dead?
- maybe….
6: Well, I love you... I will come to your grave and give you flowers.
- (why are you pointing at my head?)
6: And I will make sure you never become a zombie.
I have sometimes procrastinated on some presentation for so long that I’ve seriously contemplated faking my on death to get out of it.
But then who hasn’t?
postscript:
"But why?" you ask?
Simple. The entire system is designed to minimize false positives. Who cares about false negatives when you have 5-50 times more applicants than slots?
…and, yes, it is FAR FAR FAR worse for faculty positions. I've got numbers and everything.
Wait. If permanent daylight saving time means kids walking to school in the dark… couldn’t we just have school start later? Don’t we need to do that anyway? </stares in teenager>
While I’m saying obvious things, let me point out that surveys showing that students want to come back to campus this Fall are not showing they want to come back to *class*.
Occasionally I wake up and realize that the 1980s weren’t 20 years ago, and I find myself profoundly angry. I am currently contemplating starting a war.
Introducing a new video series with me and
@isbellHFh
. We're watching Westworld together and commenting on the AI/CS/ML ideas they touch on. We call the series "Smoov and Curly on Video"! Here's the teaser: . First episode drops tomorrow.
Look, calling a candidate diverse is objectifying and reductive, yes, but more importantly calling a *candidate* diverse makes it easier to avoid noticing that the candidate *pool* is not.
Also, it irritates me.
"Collusion Rings Threaten the Integrity of Computer Science Research" by
@MLittmanCS
@BrownCSDept
says attempts to "game" the peer-reviewed
#conference
publications system could undermine the mechanism and damage the ability to share
#research
effectively
So very happy to see that in the first ever USNWR undergrad rankings of CS,
@gtcomputing
is ranked
#5
.
There's lots of other noteworthy news in the rankings as well, including being ranked
#1
in cybersecurity!
@Freeyourmindkid
Where I was in grad school, to get licensed to cut hair, one had to prove one could cut White hair but not Black hair. It was hard to get a haircut. Eventually I just started shaving my head. Imagine being a Black woman there. There's an extra cost in even everyday things.
I'm not angry. I understand the beauty and power of LISP and am only sad that the rest of you do not.
But my Language teaches forgiveness because the expression in s-expression is love.
We have programming language wars because languages define the boundaries of communities. People look out from their community and realize that not everybody agrees with all the things about coding that their group knows deep in its bones to be true, and it makes them ANGRY.
Bittersweet news:
@OhioState
announced today that it has hired Ayanna Howard as its next Dean of Engineering. Congratulations,
@robotsmarts
& all the best in your new role!
ONE SHOULD NEVER CONFUSE ONE’S GROWING AWARENESS OF SOMETHING WITH AN ACTUAL GROWTH OF THAT THING.
thank you for coming to my all-caps evergreen ted talk.
Thread: DHH receives credit limit 20x higher than his wife for
@AppleCard
, even though it turns out his wife has a higher credit score than him. When they try to find out why, are told by Apple "I swear we're not discriminating. It's just the algorithm."
We are excited to announce that Raheem Beyah has been selected as our new dean of the College of Engineering and will begin his duties as dean on January 15. Congratulations to Dr. Beyah!
Here is your occasional reminder that doves and pigeons are the same thing.
...and you need to do a little introspection about why you want to think of them as different and how that perpetuates the inequities of the current system.
Here's my conversation with Michael Littman (
@mlittmancs
), professor of CS at Brown, about the history of reinforcement learning & the future of AI. I haven't listened to Taylor Swift, but Michael made a strong case that I should. This chat was a fun one.
Your occasional reminder that when you find yourself about to say “just” in an academic context—as in “this is just that”—you might want to surface your assumptions. You’re probably skipping a lot of steps.
@sydgibs
Both of my children’s nicknames are acronyms of their names. Those acronyms are themselves actual names that do not have natural nicknames.
The world is as it should be, at least in that respect.
Honored today, Dr. Dorothy Cowser Yancy was
#GeorgiaTech
's first African American faculty member to gain tenure. Back in the day,
#GTComputing
Dean Charles Isbell was one of her undergraduate students!
I can’t figure out which twitters I’m in.
Right now I think I’m somehow in Econ Twitter, definitely in Annoyed AI Researchers Twitter, and certainly a few others, but I don’t have good names for them.
Georgia Tech’s new Student Center will be named in honor of John Lewis, the renowned civil & human rights leader. Our campus sits in the congressional district which Congressman Lewis served for 33 years. The new facility is set to open in Summer 2022. 🐝|
@roydanroy
I hate this kind of review.
“This algorithm and result are obvious, even though no one has produced this algorithm or result, and in fact the simplicity comes from your well thought out and clear explanation. Next time please use more gratuitous math and unclear language.”
The responses to this thread pretty much sum up everything important our descendants will need to know about us in 6,000 years.
They won’t understand it at all and I’m pretty sure it will spawn multiple religions that will NOT get along.
I’m going to make a new website for my lab. How hard could that be? I’ll use Jekyll.
To get Jekyll I need to update Ruby. To update Ruby I need to update homebrew. Homebrew update fails.
@mattyglesias
@julia_azari
You could just ask me. We have this whole MOOC-based MS degree from a top 10 Computing unit that went from 0 to 10,000 students in ~6 years. We've learned lessons and everything.
When I win this random drawing, I'll probably make you read all about it as a white paper, anyway.