I successfully defended my PhD today (exciting!).
I've struggled today balancing feelings of accomplishment+gratitude and anti-climatic disappointment.
It's been more than worth it to hear my youngest daughter toddle around saying "Doc-tor Dad!" 🥰
‼️ 🎉🎊🪅🥳
It's been a really long time since I saw the words "Accept" next to a submission of mine. I'm super pumped to be able to publicly share what we've been working on for the last while. Stay tuned!!!
#NeurIPS2021
I’ve been slowly and surely making my way through
@PatrickKidger
’s thesis (it’s remarkable btw):
While I’ve been learning a lot I’ve found it hard to know which type of NDE I want to use. Does anyone know of a clear taxonomy between ODEs, SDEs and CDEs?
It's been a great Memorial Day to pause and reflect.
It's been a whirlwind season as I defended and formally graduated with my PhD
@UofTCompSci
+ moved my family across the US to California to start a postdoc
@Apple
!
Lots of exciting things to come!
🎉Last week, amid the ICML rush, we were informed that our paper (w/ Sonali Parbhoo and
@MarzyehGhassemi
): "Risk Sensitive Dead-end Identification in Safety-Critical Offline Reinforcement Learning" was accepted to
@TmlrOrg
! 🎉
Short thread introduction 👇 🧵 (1/7)
Today is my first day
@GoogleAI
as an intern! It's been a little weird to be starting a new job/adventure from my bedroom office, but I'm excited for what's to come!
This summer I'm pumped to be working alongside
@MarlosCMachado
,
@marcgbellemare
and
@pcastr
as a research intern in the Montréal
@GoogleAI
office! I've admired the work coming out of this office for a long time, I'm looking forward to getting a chance to be a part of it. 👨💻
Today I'm handing in my final assignment for my final course as a student. Capping a wild month, semester and pandemic year.
After today I'm done with all "in-person" reqs at
@UofT
and will be joining
@MIT_CSAIL
and
@MIT_IMES
as a visiting student for the remainder of my PhD!
Our paper investigating the quality of representations learned from healthcare data for use in RL will be published in the proceedings of the coming
#ML4H
workshop
@NeurIPSConf
!
w/ H. Zhang, J. Subramanian,
@mefatemi
and
@MarzyehGhassemi
Short 🧵 (1/X)
I'm humbled and thrilled to have been invited to speak at the upcoming
@OfflineRL
workshop
@NeurIPSConf
this year. It's intimidating to be the least qualified but I'll get to share how we're bringing offline RL to bear in new ways for safety-critical problems (e.g.
#ML4Health
)!
We are back for 3rd Offline RL workshop, which be held in-person at
@NeurIPSConf
2022.
The theme of the workshop is using "Offline RL as a Launchpad". Call for papers and details at .
(Canadian) Thanksgiving gratitude post! 🦃🍂
I'm grateful to be able to share our paper "Medical Dead-ends and Learning to Identify High-risk States and Treatments" (w/
@mefatemi
, J. Subramanian and
@MarzyehGhassemi
), to appear at
#NeurIPS2021
!
🧵 1/
Now that it's public-public... I'm pumped to be moving with
@MarzyehGhassemi
back "home" to New England while we continue to work toward making ML work within healthcare.
Personal news: After three awesome years at UofT CS/Medicine, my research group will be moving to MIT
@MIT_IMES
/
@MIT_CSAIL
in July 2021! Thanks to my UofT chairs and the Vector Institute for their support - excited to continue work in "healthy" machine learning for health!
🚨 New work, now public 🚨
With Haoran Zhang,
@tom_hartvigsen
and
@avapamini
, I'm pleased to share: "Continuous Time Evidential Distributions for Irregular Time Series" which I'll present tomorrow at
#ICML23
🏝️ in the IMLH Workshop () 🪡🧵
This summer I'm pumped to be working alongside
@MarlosCMachado
,
@marcgbellemare
and
@pcastr
as a research intern in the Montréal
@GoogleAI
office! I've admired the work coming out of this office for a long time, I'm looking forward to getting a chance to be a part of it. 👨💻
🚨 New work, now public 🚨
With Haoran Zhang,
@tom_hartvigsen
and
@avapamini
, I'm pleased to share: "Continuous Time Evidential Distributions for Irregular Time Series" which I'll present tomorrow at
#ICML23
🏝️ in the IMLH Workshop () 🪡🧵
I'm thrilled to announce that I have decided to attend
@UofT
@VectorInst
in pursuit of a PhD, working with
@MarzyehGhassemi
and many other future colleagues and friends!!
I am once again honored to be among those named as “Top Reviewers” for
#NeurIPS2022
It’s also gratifying to see many names of friends and colleagues also listed. It’s wonderful to see so many people recognized for doing their best job as a reviewer.
[**New work**]
We've posted a preprint of initial work looking to make transfer learning feasible in offline RL using counterfactual inference.
"Counterfactually Guided Policy Transfer in Clinical Settings" with
@MarzyehGhassemi
and
@shalmali_anu
[1/10]
We were surprised and delighted by this package today from
@Microsoft
+
@MSFTResearch
(to borrow from somewhere else). I would’ve never guessed that we’d get something like this as an intern. S/o to the reference about a 4th child.
Woke up to a nice email from the
@NeurIPSConf
chairs informing me that I was named in the top 8% of reviewers which includes free registration to the conference.
I’m happy and proud to be recognized. Attentive reviewing is among the most important contributions we can provide
It’s been an adventurous few days as we moved from Toronto 🇨🇦 back to Massachusetts 🇺🇸, & rested in Cape Ann 🏖 in preparation for an eventful summer!
Today is my first day
@Apple
where I’ll be working with the Health AI team! I’m excited for what I’ll get to learn and achieve!
I’m in Seattle this week to kick off my summer
@MSFTResearch
! Excited to be working with
@apsoleimany
to extend scalable uncertainty estimation techniques to irregular biomedical time series.
It’s been 10 years since I was last in the PNW and it’s been delightful so far!
Unfortunately ICML “won” this round. Ultimately 2/2 rejections of great papers that I’ve been working on for some time. While disappointed, I’m grateful to have the opportunity to improve them for future submissions.
@IAmSamFin
I'll put some thought into it and will get back to you. I feel like I'm always catching up and relearning everything, a perpetual novice...
One resource that I really admire is
@dustinvtran
's ml-videos repository that he's been curating:
I'm happy to share a blog post I co-authored with
@mefatemi
and
@MarzyehGhassemi
for
@MSFTResearch
, highlighting our ongoing work trying to help experts in safety-critical sequential decision making settings by recommending things to avoid doing.
An AI model trained using offline reinforcement learning could alert human decision-makers when a set of decisions or the situation is potentially high-risk (or both), in safety-critical domains such as in healthcare. Learn more about this research at:
Celebrating 10 years married to my best friend! It’s been a wild year but it’s been made far easier by having her and our little pandemic partners around to hang with. Family is what it’s all about when all is said and done, I’m fortunate to have a happy and healthy one.
I just finished my set of reviews for ICML. It's my first time reviewing for the main conference and I was struck with two major impressions: (1) There's a ton of fascinating work going on and (2) I have a greater appreciation for the helpful reviews I've received in the past 1/x
Parenting is the best example of learning in the presence of distribution shift. Even with moment matching regularization, things can still get a little silly.
🛫 On my way to
#NeurIPS2022
in New Orleans!
🤓 I’m excited to catch up with people after having been stuck in/on a computer for the past three years.
🤖 I’m looking forward to chatting about all things RL, uncertainty guided decision making, and healthcare.
See you there!
I grew up at the foot of these mountains. I took their majesty for granted. Every time I come back to visit, I find myself just staring up at them. Woke up to this view and a chorus of birdsong. Sweet comforts in a particularly tough time.
Eugene helped me find the role that I'm in
@Apple
and has set the bar exceptionally high for curiosity driven research with potential for major impact.
It's been a joy working alongside a friend these 6 months. I can't wait to see what he and his students produce
@nyutandon
!
I’m happy to announce a successful replications of
@CSProfKGD
’s recent results. RL agents can indeed generalize when given the right training conditions, time, patience and a cool new 🚲
@BlackHC
@DeepMind
@MetaAI
If you're just being rejected now, then you fared much better than I did in the application process... I received rejection notices within 24 hours.
You're doing great!
I’ve been aware of these rumblings for the last little while. I am excited for
@RL_Conference
and am looking forward to supporting it in whatever way I can. It is way past time to have a premier RL-focused venue and I can’t wait to submit my own work here in the future!
Thrilled to announce the first annual Reinforcement Learning Conference
@RL_Conference
, which will be held at UMass Amherst August 9-12! RLC is the first strongly peer-reviewed RL venue with proceedings, and our call for papers is now available: .
Having just finished writing my dissertation… this hits a little too close to home.
How do you sufficiently encapsulate everything you’ve done in 3-5+ years?! Oh, by skipping all the failed research attempts?
It is hard to encapsulate how great it was to attend
@RL_Conference
. Having been a part of the launch of 2 new venues (
@CHILconference
+
@SymposiumML4H
) in the past, it’s remarkable how well RLC went in its first iteration. A real testament to the organizers and the RL community.
I had the opportunity to be interviewed on
@TalkRLPodcast
and talk about my work in RL applied and motivated by healthcare challenges. We covered a lot of ground during the interview, it was great!
I'm interested to hear what people think and if I can further explain things.
Episode 13
@tw_killian
on the latest in RL for Health, including Hidden Parameter MDPs, Mimic III and Sepsis, Counterfactually Guided Policy Transfer and lots more!
I'm going to be spending the next week in Vancouver at
@NeurIPSConf
#NeurIPS2019
!! I'm looking forward to meeting with friends and collaborators as well as learning from my peers and colleagues. It's going to be great!
If you want to talk RL x Causal Inf x Health, pls reach out!
Running some real world RL experiments today with a few high entropy agents (some observing and one executing). This is as close to robotics that I’ve ever gotten, seems fun.
🐛 + 🧩 = 🥳
I'm looking forward to finally getting a chance to present this work tomorrow afternoon in the
#ML4H
workshop at
#NeurIPS2020
!
I'm poster 73, scheduled for Session B () but will be around the workshop all day! I'm looking forward to seeing folks!
Our paper investigating the quality of representations learned from healthcare data for use in RL will be published in the proceedings of the coming
#ML4H
workshop
@NeurIPSConf
!
w/ H. Zhang, J. Subramanian,
@mefatemi
and
@MarzyehGhassemi
Short 🧵 (1/X)
I've been extensively impressed with this paper this weekend. After poring over
@pcastr
's thread, I knew it was going to be a "must read". This is especially true for those of us looking for ways to make Offline RL plausible.
1/x
We know offline RL (e.g. fixed dataset) is hard.
What if we train on *identical data* provided by a regular RL agent? Turns out this is also hard!
Our
#NeurIPS2021
paper proposes "Tandem RL" to explore this.
Paper:
Blog:
1/🧵
Finale introduced me to the research and career path that I've happily pursued over the past few years and (I hope) well into the future. My heartiest congratulations to her well deserved promotion to full professor with tenure!
For
#ReadAcrossAmerica
our school encouraged students to come dressed as their favorite character. Our princess obsessed K-grader settled on this look. Thanks
@haleshannon
and LeUyen Pham for bringing the Princess in Black into our home. (bonus: Bat Baby wanted in on the fun)
It was a nice surprise today to get en email from
@TmlrOrg
informing me that I’ve received “Expert Reviewer” recognition.
I’m glad my efforts have been recognized. I hope that this means that I’ve “made it” within my research community, take that impostor syndrome.
There's nothing quite like deepening one's impostor syndrome by jumping into LLM research, finally getting some fine-tuning experiments running, and discovering that your resulting generations are complete garbage...
#fridaysinresearch
#feelingtheperplexity
I'm really excited to be on my way to
#RLDM
this week in Providence, RI! I've really missed in-person meetings and chances to catch-up with and meet new friends.
I'm looking forward to seeing everyone there!
I'm excited to once again be "attending"
@NeurIPSConf
with an accepted paper!
@mefatemi
and I will be presenting our "Medical Dead-Ends" work tomorrow in Poster Session 1 and are looking forward to seeing people there!
#NeurIPS2021
(Canadian) Thanksgiving gratitude post! 🦃🍂
I'm grateful to be able to share our paper "Medical Dead-ends and Learning to Identify High-risk States and Treatments" (w/
@mefatemi
, J. Subramanian and
@MarzyehGhassemi
), to appear at
#NeurIPS2021
!
🧵 1/
New achievement unlocked! My first "NFT"?
What a fun prize after submitting a review to
@JmlrOrg
.
With today's announcement of
@TmlrOrg
being open to submissions, I can only look forward to getting asked to review in hopes of getting another badge!
We’re having a great time at
#RLDM2022
! Come in through the Kaspar room in the Brown student center if you’re around. Will be here for another 90 mins or so
As a Christmas present, my 8 y.o. and I were gifted a ski day. My first time going in 10 years and his first time ever!
Hopefully we get to go again before another 10 years passes!
Ending my work day feeling grateful for remarkable collaborators who went above and beyond over the last month or so. I'm lucky to work with people who are caring and good.
#NeurIPS2023
I interned with this team and it was among the most enjoyable periods of my PhD. Everyone involved in
@MSRNE
's BioML group is kind, insatiably curious, and supportive. To top it all off, they're at the forefront of exciting advancements and are making things happen!
This is one of those papers that I wish that I had written.
@BenKompa
,
@latentjasper
and
@AndrewLBeam
do a great job outlining uncertainty quantification methods relevant to prediction in medical ML.
There's a lot to build from in this paper, I'm happy that it's been published!
New perspective in
@Nature_NPJ
Digital Medicine w/
@latentjasper
and
@AndrewLBeam
!
"Second opinion needed: communicating uncertainty in medical machine learning"
🧵 below! 1/n
I've been studying up on SSL methods and approaches this summer and found it difficult to make sense of how each subsequent paper builds on the last, etc.
@araffin2
does a really great job to clarify and streamline the developments we've seen over the last couple of years.
Although not new (Siamese net are from the 90s), Self-Supervised image representation Learning (SSL) is getting a lot of attention recently, as it gets closer to supervised learning performance.
⬇️ A thread trying to summarize recent advances and their challenges ⬇️
It was joy to get out for a run this morning on a bright, sunny Saturday after a week of rain. Look at that sky!
I was hit with a wave of gratitude to be where I am at this point in my life with a lot to learn and room to grow.
There are a lot of things that make research hard and silver linings are sometimes few and far between...
But, the satisfaction that comes after squashing a bug in your code and your ideas start to *actually* work is a pretty amazing feeling.
Another day at
#RLDM2022
and another poster talking about Dead-ends in offline RL! I'm excited to debut some of our initial attempts to ensure more robust risk estimation via distributional RL w/ Sonali Parbhoo and
@MarzyehGhassemi
Didn’t pay much thought to the date while accepting my kids eagerness to bring me lunch. Grateful feelings of surprise were quickly replaced with a bunch of mischievous giggles.
@mckaycoppins
You did a nice job capturing the melancholy that currently characterizes airports. I traveled through O'Hare last week too, albeit through the international terminal (transiting between Toronto and SLC) and it was a ghost town. In Toronto, it was even quieter.
Back to school... Now the party starts in earnest. It's been fun sharing the excitement of the season with these two, all starting at new schools. Starting Kindergarten, Grade 2 (Jackman Ave) and Grade 21 (
@UofT
@UofTCompSci
)
#phdlife
The HealthyML group is gaining some real exciting momentum right now. It’s going to be a fun week at
#ICML2023
. You should 100% seek out
@VMSuriyakumar
, Haoran +
@yang_yuzhe
to chat with them about their work!
This is today! I’m excited to be splitting time between the TS4H (poster) and the
@OfflineRL
workshops (panel + invited talk).
If you’re curious how we’ve developed offline RL sol’ns to identify states+actions to avoid, make your way to Room 291-292 this afternoon!
Friday, I’ll be participating on a panel *and* speaking at the
@OfflineRL
workshop (work w/
@mefatemi
, Sonali Parbhoo and
@MarzyehGhassemi
) as well as presenting a poster in the “Time Series for Health” workshop (work w/
@avapamini
).
Next week (Sept. 13) I'll be visiting
@NYUDataScience
to speak at their Lunch Seminar series!
I'm looking forward to another chance to share the fun I've been having repurposing Offline RL for data constrained safety-critical settings.
Squashed some bugs, submitted some experiments and am now on my way to Amherst, MA for
@RL_Conference
!
Being in New England is worth the trip on its own. To partner that with this new conference geared toward the research questions I find most stimulating makes it even better!
Seconded.
I’m recalibrating my expectations for the Spring but mostly excited to build some more legos and read more picture books. Might get crazy and build a fort for a princess tea party.
Stop assuming all academics can be more productive while required to work @ home bc of
#COVID19
. The ability to work w young children at home is zero. Yet another example of policies (however needed) having disproportionate impact on early career people (mostly women) w children.
Some of the best things I've learned from my short time working within Brain has been the astounding quality of Samy's leadership. I quickly saw that the friendly and supportive culture that I observed from the outside seems to derive from his and other senior leader's character.
Samy Bengio has stood by all of us in Brain Research, and it’s meant so much. Particularly for all the women who speak up. We started to call it “Samyland”, a safe space for the underrepresented and overlooked.
We had a great discussion about the intersection of Causal Inference and RL at PCSOCMLx
@MILAMontreal
! Thanks for everyone that came and help cover for my lack of expertise.
Today’s afternoon activity after a good week of snow ❄️
We ran out of steam as it got hard to keep building higher but had a lot of fun along the way. We’ve got plans to finish if it stays cool enough and we get more snow. 🤞🥶🤞
I’ve been anxious for this to be announced since first hearing about the idea. I’m excited about TMLR as I believe that it provides more equitable access to high quality reviewing as well as promoting good and deliberate science. I’m planning to get a paper ready for submission!
Today,
@RaiaHadsell
,
@kchonyc
and I are happy to announce the creation of a new journal: Transaction on Machine Learning Research (TMLR)
Learn more in our post:
My kids have discovered a new game.
They call it "Scare Dad!"
It's this fun development where they've learned to sneak up to my desk undetected, due to noise cancelling headphones and some moderate amount of focus, and then they pounce!
That's how today's WFH is going... 😲
To add to the overwhelming sense of "there's too much to read/attend to" we've released the accepted papers to our upcoming
@RL_Conference
workshop. These were all interesting + unique investigations into RL as a conceptual framework.
Take a look and we'll see you in Amherst!
✨We have just released all accepted papers on our website✨
Really amazing work examining the conceptual foundations of RL. Our reviewers tell us that it's the most fun they ever had reviewing. We look forward to seeing you all soon in
@RL_Conference
! 🚀
10 years later and after many previous (and current bouts) of impostor syndrome + feeling overwhelmed... I still love research.
I've switched from Fluid Dynamics to Radar to finally rest at ML/RL applied to Healthcare
#ML4H
.
It's been fun and I'm just getting started.
New
@RL_Conference
workshop announcement! I’m excited for this because it’s bringing together really thoughtful and insightful people to discuss ways we can perhaps re-frame RL to contribute functional paradigms to build innovative solutions. Check the thread 👇for more details.
🚨Announcing 🫸Finding the Frame🫷: an
@RL_Conference
workshop for examining conceptual frameworks in RL!🚨
Do you want to question formulations or practices in RL? Define new problems, challenge assumptions, or test the limits of RL paradigms?
Join us!
I'm just finishing up my first year of a PhD and am already getting messages from family + friends about faculty postings at universities near my hometown.
Thoughtful, but more than a bit premature... I need to finish my
#neurips2020
submissions first... What day is it anyway?
I've spent the past couple of days getting up to speed with MIMIC-IV () and have been blown away at how much easier the MIT LCP and PhysioNet teams have made data extraction.
Kudos and thanks
@tompollard
,
@alistairewj
, and others for the yeoman's effort!
I’m looking forward to this year’s CHIL! I’m addition to the great lineup of speakers, I‘LL be presenting some of my work that was accepted for publication w/
@shalmali_joshi_
and
@MarzyehGhassemi
! (see linked thread in the replies)
@danijarh
@_rockt
I think the same could be done by just posting a technical report on your website with accompanying blogpost? Seems like the more appropriate setting rather than posting to a preprint server designated to communicate (rigorous, I hope) scientific findings IMO.