Michael Moor
@Michael_D_Moor
Followers
3,365
Following
794
Media
319
Statuses
1,844
MD. PhD. Assistant Professor @ETH Zurich. Previously @Stanford CS.
Gallifrey
Joined April 2013
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Michigan
• 306495 Tweets
Texas
• 254937 Tweets
Never Trump
• 148974 Tweets
Tennessee
• 136809 Tweets
Arnold Palmer
• 101848 Tweets
Usher
• 79319 Tweets
Inter Miami
• 67889 Tweets
Romero
• 55021 Tweets
Gago
• 47590 Tweets
Rojo
• 39533 Tweets
Bama
• 37424 Tweets
Hulk
• 37247 Tweets
Stanton
• 32708 Tweets
Arch
• 31410 Tweets
KARIME EN PALAZZO
• 27160 Tweets
Tigre
• 27093 Tweets
Kentucky
• 23736 Tweets
Milroe
• 22273 Tweets
Beck
• 21782 Tweets
GALA EN EL AUDITORIO
• 19517 Tweets
GALA X PATY CANTÚ
• 19452 Tweets
Riquelme
• 18056 Tweets
Iowa
• 17797 Tweets
Kirby
• 16874 Tweets
Ewers
• 15644 Tweets
設営完了
• 11377 Tweets
I DID SOMETHING BAD
• 10619 Tweets
Excited to announce Med-Flamingo, a new multimodal few-shot learner specialized for the medical domain!
Last week, we uploaded the pre-print, now it’s finally live!
Paper:
Code:
Model:
A short 🧵.
1/n
11
132
469
📢 New piece published today in
@Nature
!
It was a pleasure to work with such a stellar team including Oishi Banerjee,
@ZahraShakerii
,
@hmkyale
,
@jure
,
@EricTopol
, and
@pranavrajpurkar
!
🧵below.
@Stanford
@Harvard
@Yale
@UofT
@scrippsresearch
1/n
5
91
329
2022 was wild for medical
#AI
and esp. medical foundation models (FMs).
This 🧵lists some of the standout papers from this year about this topic. Let's go!
(1/9)
#medtwitter
#AIinMedicine
#medicalAI
@NatureMedicine
4
69
225
Gave my first live talk
@Stanford
about "Topol🍩gical representation learning".
Here are the slides 🖥, created with reveal.js 🔥
#MachineLearning
#TDA
#topology
#revealjs
(1/2)
8
44
222
📢 I'm on the faculty job market for 2024! 📢
Grateful for any RTs and pointers!
My vision is to develop biomedical AI systems that exhibit generalist capabilities, see e.g.
Below a selected overview of my prior & ongoing work:
7
72
216
If you want to predict clinical phenotypes using
#MachineLearning
, check out our systematic review on ML-based
#sepsis
prediction. A THREAD with take-aways that could be relevant to
#AI
in
#healthcare
in general. (--> = hints for practitioner) 1/n
2
67
200
"Zero-shot causal learning" got accepted to
#NeurIPS2023
as a Spotlight!
🌟Story time! Featuring a 🐫, loads of 💊, causality, and👌-shot predictions.
En garde!
Title: "Zero-shot causal learning" (link ⬇️)
#MachineLearning
#AI
#causality
#AcademicTwitter
1/9
2
6
32
4
34
184
I am truly delighted to announce that I will join
@jure
's group at
@Stanford
as a
#postdoc
this March!
Excited to connect & chat about
#ML
/
#AI
for
#health
and more!
@StanfordEng
@StanfordMed
#MedTwitter
(1/4)
9
4
176
We're hiring! My new group at
@ETH_en
has the following opening:
"PhD Position in Medical AI and Foundation Models"
More details here:
Please share & repost widely!
3
38
144
@DeryaTR_
As the original poster of
@SRSchmidgall
's figure I respectully disagree that this is a final warning for clinicians to become obsolete.
I believe clinicians will be empowered by being augmented with up-to-date knowledge, guidelines and DDx ideas at their fingers' tip with
7
4
94
Excited to announce that our paper "Topological Autoencoders" has been accepted to
#icml2020
! Great team effort by
@ExpectationMax
@Pseudomanifold
@kmborgwardt
@ETH_BSSE
We propose a constraint that preserves the data space topology in the latent space 1/
3
22
85
If you care about medical
#AI
/ LLMs / MLLMs, like working with H100s - H200s and enough office space, consider reaching out!
2
9
76
Super excited to call this my new office building now!
Fellow "Gaters", if you wanna chat about
#ML
for
#health
,
#TDA
, learning with structured data (graphs, time series), let's grab a ☕! Same goes to
@StanfordMed
@StanfordEng
@Stanford
@StanfordOPA
#SiliconValley
4
0
66
Grateful for having been invited to give a talk at
@NASAAmes
!
It's mesmerizing how useful multimodal medical foundation models & OOD/transfer learning could become for spaceflight w/ rich modalities and few samples etc. Tysm for the great discussion
@RyanTScott
and team(s)!
3
4
63
#NeurIPS2020
: Please write ~10k Broader Impact statements.
Also
#NeurIPS2020
: nvm, this tool wins a best paper award:
Every tech-evangelist:
#GPT3
provides deep nuanced viewpoint
Me: GPT-3, generate a philosophical text about Ethiopia
GPT-3 *spits out factually wrong and grossly racist text that portrays a tired and cliched Western perception of Ethiopia*
(ht
@vinayprabhu
)
64
383
1K
0
7
55
Yesterday, I successfully defended my PhD Thesis! Many thanks to my friends, family, and colleagues for supporting me there, and in particular to my stellar committee
@kmborgwardt
,
@KrishnaswamyLab
, and
@jennawiens
!
#PhDone
#AcademicChatter
#phdlife
#phdchat
10
1
55
Arguably one of the most underrated skills in academic research (esp in CS/ML/AI world) is to create captivating visuals, see eg 👇
0
4
26
1
5
52
Wonderful news: A newly discovered snake has been named after Salazar Slytherin.
@AcademicChatter
0
14
48
My very first chat with
#GPT4V
for medical image interpretation (non-cherry picked).
It got 5 out of 6 images wrong, alas some were tricky.
I guess many human doctors across specialty would get them (your job is safe)
#MedTwitter
#AI
Let's go through each of them below:
1/n
4
9
47
After having co-contributed 2 papers (+1 workshop paper) to
#ICML2020
last week, I hope to have provided some evidence against the stereotype that it's impossible to do a
#PhD
while having kids (1 running, 1 installing)
#AcademicTwitter
#phdchat
@thoughtsofaphd
@AcademicChatter
4
4
43
Busy week ahead, we will present our work at
#EMNLP2023
,
#ML4H2023
, and
#NeurIPS2023
!
Looking forward to meet and chat!
0
4
39
AgentClinic: a multimodal agent benchmark to
evaluate AI in simulated clinical environments.
Exciting effort led by
@SRSchmidgall
et al.
TL;DR:
AgentClinic = a new eval of
#LLMs
/
#MLLMs
for
#medicine
beyond static QA benchmarks. Language agents need to investigate, converse,
📢 New preprint on AI agents for medicine! 📢
We introduce AgentClinic, a multimodal agent benchmark to evaluate the ability of LLMs and multimodal LLMs (MLLMs) to dialogue as doctor agents with patient agents, collect measurements, and make step-by-step decisions.
Prior work
3
8
63
0
10
40
🚨Preprint alert!🚨
Interested in
#MachineLearning
in
#medicine
and
#sepsis
?
We developed and validated a
#DeepLearning
model for predicting sepsis across 5 ICU databases from 3 countries featuring 156k ICU stays and 783 patient years worth of monitoring data.
A 🧵.
1/n
2
8
37
Thrilled to see our multi-centre
#sepsis
prediction study finally
published!
Thanks to Nicolas Bennett, Drago Plečko,
@ExpectationMax
@Pseudomanifold
, Nicolai Meinshausen, Peter Bühlmann, and
@kmborgwardt
!
@ETH_en
@ETH_BSSE
@eth_math
@PHRT_CH
4
5
37
The Lancet journals just released an entire series on pediatric sepsis with
@nelsonSpinto
, Luregn Schlapbach, et al.
As part of it, we cover AI & digital solutions for pediatric sepsis, check it out:
@TheLancet
@LancetDigitalH
0
8
38
Excited and deeply honored to be giving an invited talk about "Topological representation learning" in the
@OxUniMaths
@data_sig
DataSig seminar on September, 3rd (16:00 BST).
#AcademicTwitter
@ETH_en
@ETH_BSSE
0
8
36
2yo managed to sneek her 🦄 into my wife's office and let me tell you: this research-unicorn is getting some serious work done!
@AcademicChatter
@iamjohnoliver
#unicorn
#tech
#phdchat
3
2
34
@NoContextBrits
Technicality, but it should say: "the *longest shortest path* you can walk in the world"
Otherwise you could zigzag to make it longer. If non-shortest paths were admissible, there may not exist a longest path as each candidate path may be fractally extended.
1
0
31
🌟Story time! Featuring a 🐫, loads of 💊, causality, and👌-shot predictions.
En garde!
Title: "Zero-shot causal learning" (link ⬇️)
#MachineLearning
#AI
#causality
#AcademicTwitter
1/9
The same drug 💊 can harm one patient and save another’s life. Can we predict what will happen (AI for personalized medicine), without years of trial & error for each drug? Our work on zero-shot causal learning takes an important step towards this goal.
10
62
267
2
6
32
Elated to announce this new
#EMNLP2023
findings paper, getting us one step closer to personalizable multimodal medical AI models--here, for style-aware radiology report generation.
Great team effort led by
@benjaminbyan
and co-advised together with
@pranavrajpurkar
1/ We're getting closer to Generalist Medical AI: models interpreting medical images comprehensively, drafting full reports.
Today's progress edges us closer. Our latest work demonstrates personalizing image -> report generation to a radiologist's style with few examples.
🧵
6
83
376
2
7
30
3
7
29
Mesmerizing tutorial on Bayesian Deep Learning by
@andrewgwils
at
#ICML2020
. Talks are so much more accessible if the speaker is truly and visibly excited about the topic.
1
4
30
Achievement unlocked! 🔓
Thanks for covering
@_akhaliq
!
🤩🍾
Med-Flamingo: a Multimodal Medical Few-shot Learner
paper page:
Medicine, by its nature, is a multifaceted domain that requires the synthesis of information across various modalities. Medical generative vision-language models (VLMs) make a first step in
0
32
150
3
4
29
Another one I've heard:
"Do I really have to vaccinate all my children?"
-- "No, only the ones you want to keep.."
#medtwitter
1
3
27
Wanna iterate over a list of lists in bash, eg when submitting
@weights_biases
jobs with different configs?
This is how I do it:
#Coding
#DataScientist
#MachineLearning
2
4
28
Excited to see this paper proposing retrieval-augmented LLMs for medicine (led by
@cyrilzakka
) finally out in
@NEJM_AI
!
Our research on ‘Almanac: Retrieval Augmented LLMs for Clinical Medicine’ is finally out on
@NEJM_AI
! We demonstrate that LLMs augmented with tools (browser, calculators) outperform ChatGPT-4 & others on a novel dataset of 300+ questions assessed by board-certified physicians!
6
24
105
0
5
29
Happy to announce a new preprint together with
@Pseudomanifold
@ExpectationMax
@Jutzeler_Cathy
@kmborgwardt
@AGKBorgwardt
where we scrutinize the
#MachineLearning
based
#sepsis
prediction literature:
#MedTwitter
1/n
2
8
29
Just gave a talk
@StanfordHAI
on multimodal foundation models, great discussion, thanks for having me, Ian!
0
0
28
Interesting failure mode of
#GPT4
!
It can't play "Set", a card game that is trivial to solve with a 10-line python program.
--> it can explain the game, can abstract it, write a program to solve it, but can't actually *play* it.
Check the full convo in the chat below:
3
5
27
Was a great pleasure to contribute to
@StanfordHAI
's foundation model program, talking about multimodal foundation models!
Featuring some of our work with
@jure
,
@pranavrajpurkar
,
@EricTopol
(and more) that appeared this year in
@Nature
,
@emnlpmeeting
, &
@SymposiumML4H
!
0
2
26
Had a wonderful Q&A with
@ClementDelangue
and team from
@huggingface
at
@Stanford
today!
Thank you
@edreisMD
and David for co-organizing! Also thank you
@percyliang
for joining the Q&A!
Going to be
@Stanford
later today with some
@huggingface
team members thanks to
@edreisMD
&
@Michael_D_Moor
. Join in if you’re around!
3
3
27
2
5
25
@daniela_witten
Reviewer 1/1: This approach is quite generic and flexible. The topic sounds relevant. Accept.
1
0
26
How can we unify genomic data across the scales of biology? By building an AI Virtual Cell!🏗️🔮
In this perspective, we describe what an AI Virtual Cell is, and how to build one:
With
@_bunnech
@yusufroohani
@StephenQuake
@jure
Aviv Regev
@Tkaraletsos
4
26
97
0
4
26
Excited to see our paper covered on the
@Stanford
news front page:
Link to the post:
Thank you for great article
@coreybinns
, and the strong support by Jill Wu and
@Taylor_Kubota
!
📢 New piece published today in
@Nature
!
It was a pleasure to work with such a stellar team including Oishi Banerjee,
@ZahraShakerii
,
@hmkyale
,
@jure
,
@EricTopol
, and
@pranavrajpurkar
!
🧵below.
@Stanford
@Harvard
@Yale
@UofT
@scrippsresearch
1/n
5
91
329
0
1
25
Happy to announce that our submission "Challenging Euclidean Topological Autoencoders" was accepted at the
#neurips2020
"TDA and Beyond" workshop! Joint work with
@ExpectationMax
,
@kmborgwardt
, and
@Pseudomanifold
.
paper:
code:
1/3
1
5
23
Super excited to visit
@NASA
in Huntsville to talk about medical foundation models and
#AI
🚀 Tysm for having me Sylvian,
@RyanTScott
and team!
3
0
22
"To be here, or not to be here, that is the question."
Thank you
@PSchmiedmayer
@vishnuravi
for the great demo at
@StanfordBioX
! I see huge potential for digital health 🤩
#VisionPro
#AppleVisionPro
0
1
21
0
3
22
First, some excellent reviews explaining some key preliminaries of FMs:
🔍 self-supervised learning:
a paradigm that allows for the training of AI models w/o explicit and costly labels (huge for medical applications).
(2/9)
1
5
22
Tired of manually tracking best metrics in your
@PyTorch
script?
torch_logger.BestValueLogger()
🧐tracks best values to then 💾save them e.g. in
@weights_biases
to correlate hyperparams with the 🚀*best* loss per run.
#MachineLearning
#Coding
#Python
1
7
20
When browsing source code of an established package and finding lots of typos - but only in the comments:
#codinglife
#DataScience
#tech
@AcademicChatter
#AcademicTwitter
0
7
20
Daughter wanted to know why her finger's bleeding stopped, my try below.. happy holidays and new year 🥳
1
0
20
In our local university hospital, apparently 150,000 masks have been stolen! And this by people that mostly don't need it. If this continues, within few weeks swiss hospitals might have to conduct surgery (and treat
#COROVID19
) without masks..
#medtwitter
@AcademicChatter
1/
4
12
20
1
1
20
If you are new to
#MachineLearning
in
#Health
, e.g. for
#COVID
ー19 modeling in
#CriticalCare
, check out our review on ML-based
#sepsis
#prediction
in the
#ICU
. Many pitfalls are generic and apply to other clinical prediction targets too.
#DeepLearning
#HealthTech
#MedTwitter
Happy to announce a new preprint together with
@Pseudomanifold
@ExpectationMax
@Jutzeler_Cathy
@kmborgwardt
@AGKBorgwardt
where we scrutinize the
#MachineLearning
based
#sepsis
prediction literature:
#MedTwitter
1/n
2
8
29
0
6
20
No followers so far? Use this simple trick to make it look more impressive:
2
0
19
I once knew a joke about t-SNE, but I fear I cannot reconstruct it anymore.
#MachineLearning
@AcademicChatter
#AcademicTwitter
#phdchat
2
6
19
Mindblown! 🤯
Was struggling with a deep CNN age regression task (r^2 ~ 0),
but simply reframing it as [classification + Gaussian label smoothing + expectation over prediction tensor] seems to do the trick - from 0️⃣ to 🦸♀️!
Any literature on this?
#MachineLearning
#AI
3
0
19
Visualizing point clouds of spheres using
@3blue1brown
's marvelous manim library. Looking forward to the
@icmlconf
virtual talk!
Excited to announce that our paper "Topological Autoencoders" has been accepted to
#icml2020
! Great team effort by
@ExpectationMax
@Pseudomanifold
@kmborgwardt
@ETH_BSSE
We propose a constraint that preserves the data space topology in the latent space 1/
3
22
85
2
3
19
Fascinating: another patient cured from
#HIV
using stem cell transplantation
#HSCT
, third case worldwide. Published today in
@NatureMedicine
#MedTwitter
@EricTopol
0
1
18
🔍 multimodal AI:
Medicine is driven by various types of data that hold complementary information. The rise of multimodal medical AI promises holistic views on patients and their diseases.
(3/9)
1
5
18
Thanks
@SSB_RM
@elia_guzzi
for inviting us (
@Pseudomanifold
and me) to present some of our lab's work in
#MachineLearning
for
#Personalized
#Medicine
.
@AGKBorgwardt
Great discussion at the end! Slides will follow.
1/2
4
3
17
This is exactly what they meant with "statistical significance =/= clinical relevance" in stats class
#medtwitter
@AcademicChatter
1
3
17
Me: .. but for my thesis, I will keep my .tex clean!
Also me, few weeks into writing:
\vspace{-1.7mm} $\ $ \hskip2mm \qquad \hspace*{2.5mm}
\begin{figure}[h] [H] [H!!]
#phdchat
@AcademicChatter
@overleaf
1
0
15
Finally introducing do-calculus to my little one!
#Causality
@yudapearl
#MachineLearning
#StayAtHome
This toy is excellent for demonstrating "correlation vs causation"
1
2
16
Absolutely lovely experience this
#NeurIPS2023
! Met so many AI rockstars for the first time in real life like
@iScienceLuvr
@jeremyphoward
@EMostaque
@SanhEstPasMoi
@marinkazitnik
+ many more!
Mindblowing week w/ wild 🐊, beignets and lots of music🎺🎵!
1
1
16
If you need to merge a
#LaTex
project into a single .tex file, e.g. for a paper submission, this wonderful package does the job:
$ latexpand main.tex > big_main.tex
@Pseudomanifold
@AcademicChatter
#AcademicTwitter
#ScienceTwitter
#tech
1
0
16
Looking forward as well!
Going to present topological representation learning, joint work with
@ExpectationMax
@Pseudomanifold
@kmborgwardt
@AGKBorgwardt
#MachineLearning
#Topology
#AcademicTwitter
Tomorrow (3 Sept) we're looking forward to Michael Moor's talk on 'Topological representation learning'. Find out more here
0
3
3
0
5
16
@tim_sainburg
@leland_mcinnes
Looking forward to read this!
Very interesting to see that UMAP-AE finds a similar longitudinal structure in CIFAR10 as we found with Topo-AE:
0
1
16
Interested in doing a postdoc at ETH Zurich in my group focusing on medical AI / medical foundation models?
Check out the SNF postdoc fellowship:
If interested, please check eligibility and send a brief message with CV + short 1/2 page project proposal
0
4
16
Bad stats joke of the day:
Two dependent samples walk into a bar and order the same drink.
Bartender: Sorry, but I need to see i.i.d.
#dadjokes
#statstwitter
#MachineLearning
@AcademicChatter
1
2
16
If you like this pic, you will *love* the ham sandwich theorem:
#AcademicTwitter
@AcademicChatter
RT if you agree
the transient conformation of the standwich is only stable for a few nanoseconds
39
94
1K
2
1
16
Mindblowing
#ML4H2023
experience today in New Orleans🎺
Made my day when an attendee shows me an image of how their adapting Med-Flamingo in their hospital 😭🤩
Excited for
#Neurips2023
next!
1
1
16
@iScienceLuvr
Absolutely! there are few early works already going into interactive medical LLM eval beyond MCQA:
Craft-MD:
AgentClinic:
Med-Align:
1
3
15
Announcing a small
@PyTorch
-based package "batchdist" which I created after repeatetly encountering the same issue with certain distances/losses not "really" allowing for batched computations:
#DeepLearning
#MachineLearning
#coding
1/4
1
5
14
#ICML2020
attendees, check out our recorded talk:
(open access clip will follow)
Code:
Looking forward to chat with you at
@icmlconf
!
#MachineLearning
#DeepLearning
#AI
2
3
15
One thing I love about
@Stanford
:
Today I learned (~5 min ahead) that
@karpathy
would give a Q&A next door.
Was an inspiring discussion round, involving
#AI
embodiment to break loose from social media-type data that current large models default to.
#Tesla
#MachineLearning
0
3
15
I'm so happy for you (actually both of you!) Heartfelt congratulations to your appointment, great catch by
@UCBerkeley
!
1/ So excited to announce that I’ll be joining
@UCBerkeley
in July 2025 as an Assistant Professor, with a joint appt in EECS & Computational Precision Health! Proud to be joining the school that welcomed my parents as students when they came to the US more than 35 years ago.
87
36
1K
1
0
15
Work presented was done together w/
@Pseudomanifold
,
@ExpectationMax
,
@EdwardOnBrew
, Yves Moreau, and
@kmborgwardt
.
@ETH_BSSE
@ETH_en
Also thx to
@avt_im
for recommending reveal.js, was quite a refreshing experience after years of defaulting to LaTeX/beamer. 🤝
(2/2)
1
1
13
We had a splendid poster session yesterday! Thanks for stopping by and the great discussions! 🤩
Loved how one person who went: "Zero-shot? That's impossible!🤯"
#NeurIPS2023
@h_nilforoshan
0
1
14