Lukas Muttenthaler
@lukas_mut
Followers
830
Following
4K
Media
49
Statuses
1K
Senior Researcher @AIgnostics & @ExplainableML | ex-@GoogleDeepMind | PhD in ML from @TUBerlin @bifoldberlin @MPI_CBS | Representations in 🧠 & 🤖| #FirstGen 🌈
Joined May 2019
🥳I am incredibly humbled and grateful to share that our work, "Aligning machine and human visual representations across abstraction levels," has been published today in @Nature: https://t.co/Zhjih0CSKt For more information see @AndrewLampinen’s post below ⬇️
nature.com
Nature - Aligning foundation models with human judgments enables them to more accurately approximate human behaviour and uncertainty across various levels of visual abstraction, while additionally...
What aspects of human knowledge do vision models like CLIP fail to capture, and how can we improve them? We suggest models miss key global organization; aligning them makes them more robust. Check out @lukas_mut's work, finally out (in @Nature!?) + our new blogpost! 1/4
1
7
54
I am not sure I understand the decision made by @iclr_conf to revert the scores to their state before the discussion period. That defeats the whole purpose of the rebuttal. What’s the hope there?
0
0
1
We need much more AI fairness, alignment, and safety efforts!
0
0
0
It is shocking to see that AI is so blatantly being misused for political purposes. I really hope the other tech companies won’t follow lead and ensure that responses remain fact-based and not fuelled by (right-wing) political ideology like Grok is. This is a dangerous trend.
Some of the political bias in AI models is absurd.
1
0
3
What aspects of human knowledge do vision models like CLIP fail to capture, and how can we improve them? We suggest models miss key global organization; aligning them makes them more robust. Check out @lukas_mut's work, finally out (in @Nature!?) + our new blogpost! 1/4
8
42
289
See here for the @GoogleDeepMind blogpost about our work:
Humans think about things conceptually – like how cats and starfish are both animals, despite their differences. But AI can sometimes miss this nuance. 🖼️ Our research teaches vision models to better organize visual concepts, making them more reliable and better at generalizing.
0
0
1
Humans think about things conceptually – like how cats and starfish are both animals, despite their differences. But AI can sometimes miss this nuance. 🖼️ Our research teaches vision models to better organize visual concepts, making them more reliable and better at generalizing.
56
181
1K
🧵🎉 Our mega-paper is finally published in TMLR! We're "Getting Aligned on Representational Alignment" - the degree to which internal representations of different (biological & artificial) information processing systems agree. 🧠🤖🔬🔍 #CognitiveScience #Neuroscience #AI
5
37
151
Our perspective on representational alignment is finally published in TMLR! It’s a years long collaboration between several researchers in the fields of machine learning, cognitive science, and neuroscience! Special shootout to @sucholutsky for co-leading this effort with me 🙏🏼🎉
🧵🎉 Our mega-paper is finally published in TMLR! We're "Getting Aligned on Representational Alignment" - the degree to which internal representations of different (biological & artificial) information processing systems agree. 🧠🤖🔬🔍 #CognitiveScience #Neuroscience #AI
0
0
9
Our democracy is not self-executing. It depends on us all as citizens, regardless of our political affiliations, to stand up and fight for the core values that have made this country the envy of the world.
24K
32K
201K
So much disinformation on this site tonight in the wake of the horrific Kirk murder, with conservatives cynically suggesting that all of the political violence in this country comes from the left. In fact it’s the opposite. Here are the facts for those with short memories: The
2
18K
83K
JD “I don’t give a shit” Vance says killing people he accuses of a crime is the “highest and best use of the military.” Did he ever read To Kill a Mockingbird? Did he ever wonder what might happen if the accused were immediately executed without trial or representation??
Killing cartel members who poison our fellow citizens is the highest and best use of our military.
35K
16K
146K
The President of the United States is deploying the military onto US streets and using our troops like political pawns. DO NOT ALLOW YOURSELF TO BECOME NUMB TO THIS.
10K
16K
88K
Over the long term, we shouldn’t have political gerrymandering in America, just a fair fight between Republicans and Democrats based on who’s got better ideas. But since Texas is taking direction from a partisan White House and gerrymandering in the middle of a decade to try and
We’re grateful to President @BarackObama for being in this fight with us and supporting our mission. California is advancing a process that gives voters the final say on implementing new maps. This is a fair and responsible response to Trump’s unprecedented power grab.
22K
27K
178K
We can’t lose focus on what matters – right now, Republicans in Texas are trying to gerrymander district lines to unfairly win five seats in next year’s midterm elections. This is a power grab that undermines our democracy.
Texas: Your vote matters—and so does your map. And right now, power-hungry politicians are trying to gerrymander it to silence your voices. Join @RedistrictFdn for a virtual event to learn how redistricting shapes your vote—and how you can take action. 🔗 https://t.co/FvblbIVmPw
27K
27K
140K
In neuroscience, we often try to understand systems by analyzing their representations — using tools like regression or RSA. But are these analyses biased towards discovering a subset of what a system represents? If you're interested, check out our new commentary! Thread:
5
64
363
How can we circumvent data scarcity in the time series domain? We propose to leverage pretrained ViTs (e.g., CLIP, DINOv2) for time series classification and outperform time series foundation models (TSFMs). 📄 Preprint: https://t.co/3at0eaVsEL 💻 Code: https://t.co/GbmqJUuQyI
5
21
61
We find self-supervised & language-aligned ViTs scored highest on CSS — even matching humans. Supervised models are not even close. Surprisingly, high ImageNet accuracy does not guarantee high configural shape score! 5/15
2
4
52
"On the Ability of Deep Networks to Learn Symmetries from Data: A Neural Kernel Theory" now accepted at JMLR! 🥳 🔗 https://t.co/QNrn7bl4oJ We thank the reviewers for expert suggestions which allowed us to substantially improve the work and writing. See ⬇️ for more info and 🧵
Now accepted at JMLR, and with an extension to general finite groups (including non-abelian groups)! Updated version of our (w/ @StphTphsn1) work: https://t.co/JPW4orZDKB
4
42
230
Independence Day is a reminder that America is not the project of any one person. The single most powerful word in our democracy is the word ‘We.’ ‘We The People.’ ‘We Shall Overcome.’ ‘Yes We Can.’ America is owned by no one. It belongs to all citizens. And at this moment in
30K
38K
298K