Sr Principal Researcher, Microsoft Research, NYC. Machine learning transparency, intelligibility, fairness. WiML co-founder. NeurIPS 2021 Program Co-Chair.
I want to talk about burnout. A brief 🧵...
I was well aware I was burnt out in the fall, but it's hard to fully appreciate the impact of burnout in the moment.
After 2 weeks of vacation and a month of aggressively blocking daily focus time, the impact has become more clear:
Got back reviews complaining our paper lacked novelty and didn't engage with the literature because we didn't cite... the arXiv preprint of the same paper. This is a new one for me.
The NYU AI School is a FREE week-long winter school on AI and ML open to undergrads from any major—programming experience helpful but not required.
There will be hands-on labs and introductory lectures. I will speak on Responsible AI.
Check it out!
I've seen several tweets this week suggesting that newcomers stay away from ML research unless they already have a killer idea.
Please don't take this advice.
Almost no one starts PhD research with a clear idea of what they'll do, and those who do often change directions. (1/3)
Ok, people! Are you looking for something to read at the intersection of machine learning and HCI?
Three new papers posted online today that you should check out, all with my amazing colleague/BFF
@hannawallach
!
Ready? I'm gonna try a thread!
I'm partway through reading the latest
@NIST
report on managing bias in AI and already my neck is starting to ache from the amount of vigorous head nodding I've done. So far, it is really, really good—and covers so much more than bias mitigation. Read it!
My fully-vaccinated husband came down with a fever on Friday.
On Saturday he tested positive for COVID.
The past few days he's had headaches, muscle aches, fatigue, appetite loss, a cough... the whole gamut of COVID symptoms. But it could be worse. [1/n]
A Ph.D. student collaborator of mine with a degree in CS just asked me what emacs is and I nearly fell out of my chair. I have never felt so old in my life! I still have tears in my eyes. 😭😂
Calling all PhD candidates interested in fairness, accountability, transparency, and ethics of AI!
Come work with me,
@hannawallach
,
@katecrawford
,
@haldaume3
, Miro Dudik, and the rest of the Microsoft Research NYC FATE group this summer!
Apply now:
This week marks my 10-year anniversary at
@MSFTResearch
NYC. I've been trying not to get all sentimental, but in terms of major life decisions, well, joining MSR was a good one.
#NeurIPS2019
workshop decisions just went out!
Want to hear how decisions were made and see a preliminary list of accepted workshops?
@MarzyehGhassemi
,
@shakir_za
, Bob Williamson, and I wrote a blog post all about the selection process!
Why am I sharing this?
We're still in a pandemic. The pace of tech—particularly AI—is relentless. Everyone's overextended. I'm not the only one who's been burnt out.
We need to recognize the impact on ourselves and our colleagues and find a way to make space to recover. (5/5)
As a researcher in responsible AI in industry, I have enormous respect and admiration for
@timnitGebru
,
@mmitchell_ai
, and their colleagues—for asking important questions, the diverse team they've built, and their perseverance in pushing for impact (e.g., with model cards). 1/2
It's that time of year! The FATE group at
@MSFTResearch
NYC is now accepting applications for 2023 interns. Internships this year will be IN PERSON again at last! 🥳🎉
For full consideration, apply by December 1 and don't forget your cover letter.
My 2010s:
became mom to 2 incredible daughters
bought a home
started faculty job; hated it; quit
started real dream job
mentored awesome people
wrote a bunch of papers
ran 4 half marathons
took up yoga
lost 2 cats
shifted research focus/methods
gained confidence; spoke up more
- I am less angry. This is huge.
- I feel less like every situation is adversarial, more willing to give people the benefit of the doubt.
- I feel less threatened by other people's actions, more in control of my own outcomes.
- I feel optimistic about those outcomes. (2/n)
Submitting to
#NeurIPS2022
? Want to include a thoughtful reflection on the limitations of your work but not sure how to begin?
We designed REAL ML, a set of guided activities to help ML researchers recognize, explore, and articulate limitations that arise in their research.
My favorite part of
#NeurIPS
is catching up with old friends over drinks in hotel bars followed by long, drawn-out dinners. Ducking out mid-tutorial to cook and put my kids to bed is somehow not the same experience.
I hope your managers are telling you this, but it's ok to be distracted from work right now. We're all distracted from work. I would worry about you if you were not distracted from work. Nobody is going to look back on this week in US history and regret not just doing their job.
I’ve been silent because it’s not clear that America needs to hear from another white lady, but it also feels cowardly to say nothing. Black lives matter. I am listening, continuing to learn, figuring out how to make donations count most, and trying my best to offer support.
Mark your calendar now!! On top of eight keynotes,
#NeurIPS2021
will feature three plenary panels:
- The Consequences of Massive Scaling in ML
- The Role of Benchmarks in the Scientific Progress of ML
- How Should an ML Researcher Think about AI Ethics?
I'm chronically behind on paper reading, but trying to catch up on recent lit on human-AI decision making—complementarity, under/overreliance, explanations, all that.
There is SO MUCH happening in this space, so help! What are favorite recent papers I should definitely not miss?
I failed to block my calendar for
#CHI2021
and have proved once again that I am incapable of attending virtual conferences without marking myself out of office, cancelling all my meetings, and making a serious mental commitment in advance to go all in. How do people do it?
Spread the word! 📣
We are conducting a paid study on the societal impacts of assumptions in machine learning research! Looking for ML researchers, ML practitioners, HCI researchers who work with ML, or experts in the implications of ML.
Sign up here:
My husband gave me every introvert parent's pandemic dream gift of 2 nights out of town, alone, with nobody I have to talk to, no remote school days, and nobody to look after but myself. I have never needed solitude so badly in my life and it exceeded my wildest expectations.
It's that time of year again! The FATE group at
@MSFTResearch
NYC is accepting applications for 2024 interns. 🥳🎉
We'll begin reviewing applications in early December. Cover letter required; see the posting for details.
I’ve hit a whole new level of burnout and exhaustion the past few weeks. I try to stay strong all the time but sometimes I’m not sure how we are going to get through this.
Seeing lots of industry research internship talk.
Yes, you can email researchers you want to work with!
To stand out, demonstrate you've taken the time to understand what they do, and lay out how what you bring to the table is relevant to them.
Generic emails get you nowhere.
So excited I can FINALLY share our new work on machine learning practitioners' data documentation perceptions, needs, challenges, and desiderata, which will appear in
#CSCW2022
!
Joint work w/
@AmyHeger
,
@lizbmarquis
,
@mihaela_v
, and
@hannawallach
1/n
I don't want a future in which the only people entering ML are those with the hubris to think they've got it all figured out.
I'd rather encourage newcomers who have the patience, grit, and humility to approach an active field with an open mind and willingness to learn. (2/3)
2022 marks a decade since I quit a (particularly toxic) tenure-track position and moved to industry.
It is no exaggeration to say that in that decade I have never once felt even a momentary hint of regret about that decision—nothing but sheer, unadulterated relief.
- I can see paths forward where there were only roadblocks before.
- I can brainstorm and get excited about ideas.
- I can focus on what people are saying in meetings and feel less of a need to (poorly) multitask.
- As a result, I'm more interested in talking to people. (3/n)
Interested in learning about multi-armed bandits? My
@MSFTResearch
colleague Alex Slivkins just published an introductory book and the whole thing is available for free on arXiv.
Yesterday I saw my (fully vaccinated) parents for the first time since 2019, and this morning I booked an appointment for my own vaccine! 🥳🎊🎉 I'd forgotten what it's like to feel hopeful.
This year one of my goals is to clear out my "research debt" (a term I learned from
@hannawallach
) — those lingering projects that I believe in and want to get out in the world, but are stalled for various reasons (usually "more urgent" priorities of mine/my collaborators). (1/4)
.
@MSFTResearch
NYC is hiring a postdoc whose work bridges machine learning and any of our other focus areas: FATE, computational social science, econ-CS, or AI for systems. 📢📢📢
Is that you? Come work with us!
(FATE-specific postdoc ad coming soon)
I do agree that now is a great time to get into interdisciplinary work though!
There is so much demand for ML + health, ML + biology, ML + social sciences, and on and on. These are exciting areas and are only going to grow. (3/3)
YES! Coming from an ML theory background, I can't emphasize enough how eye-opening I've found it to collaborate with and learn from researchers trained in qualitative methods. It's changed my view of not just how we should solve problems but WHICH problems we should solve. (1/3)
Sometimes there's that one project you just love, the one you go to first when someone asks what you're working on, the one you find a way to mention in every talk you give, the one you believe in deeply regardless of what any reviewer or skeptic might say. 1/6
I recently received an email from someone who had made a Wikipedia page about me for a class project. I know I should play it cool and tons of researchers and academics are on there and it's no big deal, but at the same time... THIS IS SO COOL! 😍😍😍
For the past few years I've kept a list of everything (talks, committees, events, etc.) I've said no to. Looking back, there's not a single one I regret. But I regret having committed to do things all the time.
I need to keep reminding myself of this and up how much I say no.
Today
@Microsoft
shared the Responsible AI Standard, our internal playbook for developing and deploying AI systems. This was a multi-year collaboration involving SO MANY people across roles, disciplines, and teams. While it's still evolving, I'm proud of how far we've come.
Microsoft is advancing AI with ethical principles that put people first. Our Responsible AI Standard takes a thoughtful approach, moving from principles to practices and collaborating with industry, academia and regulators for meaningful progress.
More opportunities to join us at MSR NYC!
Our interdisciplinary lab is hiring researchers and postdocs in FATE, computational social science, economics & computation, systems, and ML.
Check out our postings! (And potential interns, check back soon!)
But I want people to know that breakthrough infections are real. And without collecting the data, we can't know how widespread they are.
Let's not pretend that things are normal, because they're not.
Please stay safe and give your loved ones an extra hug. [8/8]
Spread the word! 📢
The
@MSFTResearch
Undergraduate Research Internship program is accepting applicants! 🎊🎉
MSR is looking for candidates who are passionate about technology, offer diverse perspectives, and want to work on state-of-the-art research.
Help us spread the word! 📢
@mihaela_v
, Rich Caruana, and I are looking for a *virtual* summer intern with experience in visualization + machine learning interpretability.
Still time to apply! We'll be reviewing applications through early January.
Suggested New Years resolutions:
- Say no more often, especially to things you're only doing for a CV line/to impress people.
- Use your (relative) power to pull others up the ladder behind you.
- Speak up for people in less privileged positions whenever you can.
I'm planning a 12-week sabbatical late summer/early fall and plan to disconnect from work as fully as possible. Sharing because 1) it makes me more likely to do it, and 2) taking so much time off is scary since I don't see examples of people doing it. Maybe I can be that example.
It's been 5 months since I tweeted about burnout. I'm happy to say I've kept my burnout level low. 🥂
Since my original thread seemed to resonate, I thought I'd share some things that have made a difference for me, acknowledging I'm lucky to have an unusually flexible job. (1/n)
I want to talk about burnout. A brief 🧵...
I was well aware I was burnt out in the fall, but it's hard to fully appreciate the impact of burnout in the moment.
After 2 weeks of vacation and a month of aggressively blocking daily focus time, the impact has become more clear:
Woohoo! My survey on crowdsourcing for an ML audience is finally out in JMLR! Check it out if you're interested in incorporating crowdsourcing into your own research.
Making Better Use of the Crowd: How Crowdsourcing Can Advance Machine Learning Research
Just since Friday 3 people have independently emailed me to compliment my work. It feels like I’ve entered an alternate academic universe where people are REALLY NICE, and I like it.
Let’s all be nice. Is there a paper or talk you’ve learned a lot from lately? Tell the authors!
- My priorities have become more clear.
- Day-to-day tasks like writing have become easier to accomplish and do well.
- Because of all these things, and because I'm generally less overwhelmed, I'm better able to provide support for my collaborators and mentees. (4/n)
I first attended
#NeurIPS
during the second year of my PhD. I've considered it my home conference ever since.
This surprises people because it's not the conference I've published at most and much of the research I've done over the years hasn't been pure ML (or ML at all). 1/n
A tip for postdoc/intern hiring season: If you're emailing someone about a position, dozens of others are too. To stand out, show that you've taken the time to understand the person/lab you're emailing. Explain why you're excited and how working together would benefit you both.
As an American who's worked in the US in STEM my whole career, I can't think of a single colleague who's ever suggested we'd be better off without our students, mentors, and collaborators from around the world. It is their presence that makes our community vibrant and innovative.
The past month I've aggressively said no to meetings and put that time into wrapping up lingering old projects. It has been everything I dreamed it could be. ❤️
After spending summer/fall in back-to-back calls broken up only by extra pandemic childcare, I can't do it anymore.
Ok, people, it's happening! My sabbatical starts end of next week!!!!!! 🤯🍾🙌
I plan to be as close as possible to 100% away from work. As a first step, I'm starting my social media hiatus TODAY.
See you all in a few months if this site still exists?? 👋
I'm planning a 12-week sabbatical late summer/early fall and plan to disconnect from work as fully as possible. Sharing because 1) it makes me more likely to do it, and 2) taking so much time off is scary since I don't see examples of people doing it. Maybe I can be that example.
One of the things I've been most grateful for during the pandemic is having supportive colleagues who get that there's more to life than work, that people have responsibilities outside their jobs, and that number of hours worked is not a particularly useful measure of anything.
Microsoft Research NYC is now hiring research interns for the spring!
Interested in fairness in ML and have a background running human subject experiments or user studies? Come work with me and
@hannawallach
!
Tell your best PhD students to apply!
So excited to finally be able to share a working draft of our paper on industry practitioners' needs around fairness in ML systems!
Feedback welcome! We're prepping the final version for
#CHI2019
.
with
@d19fe8
,
@haldaume3
, Miro Dudik, and
@hannawallach
I'm passionate about research and get excited over new ideas, but how good/bad I feel about work on a day-to-day basis is more connected to the quality of my interactions with colleagues. Are they kind? Are they listening to me? Do they value my opinions, even if they disagree?
@hannawallach
#2
Our CHI 2020 paper on "Interpreting Interpretability"
Developing interpretability tools is not enough! Why user-centric evaluation of interpretability tools is necessary...
With
@harmankkaur
,
@hannawallach
, and other awesome non-twitter people.
Hey PhD students working at the intersection of AI and HCI!
MSR is hiring multiple interns for this upcoming SPRING!
Work with Saleema Amershi,
@ecekamar
,
@besanushi
,
@hannawallach
, myself, and other awesome people on human-AI interaction.
It’s been a decade since I finished my PhD but I had a visceral reaction to this post. I have strong memories of how depressing the holidays were when I was in my late 20s and still a student. I wish I’d stopped comparing myself to others and enjoyed the end of the PhD more.
Grad students, this time of year can mean extra intense comparison to friends or family your age who took jobs right out of college. Remember that their path is not your path. Stay focused on your long-term goals and the things that you are proud of having accomplished.
#PhDchat
Assessing AI systems for performance disparities between groups is harder than it seems.
In new work to appear at
#AIES2021
, we dig into the choices that must be made when designing disaggregated evaluations, and the impact of these choices.
Draft here:
Yesterday, quite unexpectedly and without planning, I broke my half personal record by almost four minutes. At age 43. There may have been some tears at the finish line.
When I think of Central Park I'll think of this. 🏃♀️
(Six months til
@nycmarathon
!)
#womenruntheworld
@nyrr
Ok, so this is full-on shameless self promotion, but I love the way that this blog post tells the story behind the work that
@hannawallach
and I have been doing with our FATE/Aether colleagues for the past few years. It makes me feel energized all over again!
For
@jennwvaughan
and
@hannawallach
, responsible AI means putting people first. In two
#CHI2020
papers, they and their co-authors aim to support developers with a fairness checklist and study of the use and perception of interpretability tools:
#MSBuild
I don’t know the details of this story, but I know
@timnitGebru
has always been a force of nature driving change in the ML community and in responsible AI practice. I'm grateful for her research contributions, community leadership, and tireless efforts to do the right thing.
The FATE group at MSR-NYC still has a couple summer intern openings for PhD students with an interest in responsible AI and experience with semi-structured interviews, co-design, and HCI methods generally.
Interested? Email me and
@hannawallach
with your research background!
Well! That was one of the most exhilarating, inspiring, and truly magical days of my life. Nobody puts on a party like NYC. ❤️ Grateful to everyone who made this experience possible — particularly the throngs of strangers screaming my name miles 24–25 in Central Park. 🏃♀️🏃♂️🎉🙌
Thanks for the support. My husband's fever lingered 4–5 days, but that and other symptoms have now faded. I can't imagine how terrifying this all would be if he were unvaccinated.
Kids and I tested negative again yesterday. Too soon to declare victory, but for now we're ok.
My fully-vaccinated husband came down with a fever on Friday.
On Saturday he tested positive for COVID.
The past few days he's had headaches, muscle aches, fatigue, appetite loss, a cough... the whole gamut of COVID symptoms. But it could be worse. [1/n]
I'd hoped conferences/seminars going remote would lead to more speaker variety since people (e.g., with kids) who can't travel much can give more talks online. Now I'm worried it will lead to less variety, since "rock stars" can be at multiple events at once. What are you seeing?
My daughter's 2nd grade class devoted the first week of math to replacing fixed mindset thoughts with growth mindset statements. As a PhD in a mathematical field with a daughter with stubborn perfectionism LIKE MYSELF, I'd like to high-five whoever came up with this curriculum.
People always ask me if I like Twitter and I always say no without thinking.
This certainly used to be true.
I don't like being bombarded with everyone's achievements, especially the rich-get-richer 37-under-37 popularity contests. I don't need a reminder I'm not popular. (1/4)
Useful perspective for research. Novelty is overrated; refining, communicating, and validating ideas are underrated.
Researcher evaluations in industry labs get at some of this—impact is often as/more important than publications. Would like to see more incentive for scholarship.
Follow-up:
@o_saja
& others brought up that it can take more than vacation and focus time to recover. 💯 I'm not fully recovered now, and I've had other burnout periods that required major life changes (like leaving academia a decade ago).
But I do think these things help...
This week at
#HCOMP2021
,
@elmelis
presents our paper "From Human Explanation to Model Interpretability: A Framework Based on Weight of Evidence" (w/
@harmankkaur
,
@haldaume3
,
@hannawallach
).
We wanted to develop explanations for ML predictions that are meaningful to humans. 1/n
🚨🚨🚨
The FATE group at
@MSFTResearch
NYC is hiring a postdoc whose work focuses on the societal implications of AI&ML!
📢📢📢
Does that sound like you? Come join us!
Group:
Job Ad:
Happy to answer any questions you may have!
Sad I won't be in Hawaii next week for
#CHI2020
, but excited that the Fair & Responsible AI Workshop has moved online and the list of short papers looks phenomenal!
I'm struggling to make sense of what is happening at Google. I keep thinking I must be missing some key piece of the story because it just doesn't add up. But I know that
@timnitGebru
and
@mmitchell_ai
will land on their feet and come out of this stronger. 2/2
Anecdotally I've found women in tech say yes to too much service, partly to prove we're part of communities where it's hard to gain acceptance.
After severe service burnout I'm stepping way back starting this month and it's amazing. I have so much more time.
Why do we do this?
Friends in the tech industry, this is such a tumultuous and dizzying time. There's momentum and pressure to charge ahead, but sometimes what you need is a few hours/days to step back and breathe. (I've been needing extra sleep just to process thoughts.) Hope you all are doing ok.