Vlad Niculae
@vn-ml.bsky.social
Pinned
Vlad Niculae
@vn-ml.bsky.social
· Apr 24
Vlad Niculae
vene.ro
Recruiting a PhD candidate at U. of Amsterdam (funded, 4yr). We will use ML&NLP, prob. models, and user studies, to make adaptive scientific-assistant systems that communicate & justify decisions in ways helpful to experts.
More: vene.ro/jobs.html
Apply by May 18: werkenbij.uva.nl/en/vacancies...
More: vene.ro/jobs.html
Apply by May 18: werkenbij.uva.nl/en/vacancies...
Reposted by Vlad Niculae
Your data is low-rank, so stop wasting compute! In our new paper on low-rank thinning, we share one weird trick to speed up Transformer inference, SGD training, and hypothesis testing at scale. Come by ICML poster W-1012 Tuesday at 4:30!
New guarantees for approximating attention, accelerating SGD, and testing sample quality in near-linear time
July 14, 2025 at 6:29 PM
Your data is low-rank, so stop wasting compute! In our new paper on low-rank thinning, we share one weird trick to speed up Transformer inference, SGD training, and hypothesis testing at scale. Come by ICML poster W-1012 Tuesday at 4:30!
Reposted by Vlad Niculae
So you want to skip our thinning proofs—but you’d still like our out-of-the-box attention speedups? I’ll be presenting the Thinformer at two ICML workshop posters tomorrow!
Catch me at Es-FoMo (1-2:30, East hall A) and at LCFM (10:45-11:30 & 3:30-4:30, West 202-204)
Catch me at Es-FoMo (1-2:30, East hall A) and at LCFM (10:45-11:30 & 3:30-4:30, West 202-204)
Your data is low-rank, so stop wasting compute! In our new paper on low-rank thinning, we share one weird trick to speed up Transformer inference, SGD training, and hypothesis testing at scale. Come by ICML poster W-1012 Tuesday at 4:30!
New guarantees for approximating attention, accelerating SGD, and testing sample quality in near-linear time
July 19, 2025 at 7:04 AM
So you want to skip our thinning proofs—but you’d still like our out-of-the-box attention speedups? I’ll be presenting the Thinformer at two ICML workshop posters tomorrow!
Catch me at Es-FoMo (1-2:30, East hall A) and at LCFM (10:45-11:30 & 3:30-4:30, West 202-204)
Catch me at Es-FoMo (1-2:30, East hall A) and at LCFM (10:45-11:30 & 3:30-4:30, West 202-204)
Reposted by Vlad Niculae
new working paper! we (me, Su Lin Blodgett, @ninamarkl.bsky.social) examine how recent marketing of LLMs extends older discourses that cast workers as bundles of skills, and unpack the false promises of empowerment these discourses embed, in times of precarity
tisjune.github.io/papers/aarhu...
tisjune.github.io/papers/aarhu...
tisjune.github.io
June 27, 2025 at 7:41 PM
new working paper! we (me, Su Lin Blodgett, @ninamarkl.bsky.social) examine how recent marketing of LLMs extends older discourses that cast workers as bundles of skills, and unpack the false promises of empowerment these discourses embed, in times of precarity
tisjune.github.io/papers/aarhu...
tisjune.github.io/papers/aarhu...
Reposted by Vlad Niculae
Looking forward to this year's edition! With great speakers: Ryan McDonald Yulan He @vn-ml.bsky.social @antonisa.bsky.social Raquel Fernandez @annarogers.bsky.social Preslav Nakov @mohitbansal.bsky.social @eunsol.bsky.social Marie-Catherine de Marnefffe !
📢 10 Days Left to apply for the AthNLP - Athens Natural Language Processing Summer School!
✍ Get your applications in before June 15th!
athnlp.github.io/2025/cfp.html
✍ Get your applications in before June 15th!
athnlp.github.io/2025/cfp.html
June 6, 2025 at 9:10 AM
Looking forward to this year's edition! With great speakers: Ryan McDonald Yulan He @vn-ml.bsky.social @antonisa.bsky.social Raquel Fernandez @annarogers.bsky.social Preslav Nakov @mohitbansal.bsky.social @eunsol.bsky.social Marie-Catherine de Marnefffe !
Reposted by Vlad Niculae
my lab (lacns.github.io) at @mpi-nl.bsky.social and @dondersinst.bsky.social is recruiting for two PhD and two postdoctoral positions funded by an @erc.europa.eu Consolidator - come join us!
PhD: www.mpi.nl/career-educa...
Postdoc: www.mpi.nl/career-educa...
(please share widely)
PhD: www.mpi.nl/career-educa...
Postdoc: www.mpi.nl/career-educa...
(please share widely)
Language and Computation in Neural Systems
We are an international group of scientists consisting of linguists, cognitive scientists, cognitive neuroscientists, computational neuroscientists, computational modellers, computational scientists, ...
lacns.github.io
May 20, 2025 at 1:49 PM
my lab (lacns.github.io) at @mpi-nl.bsky.social and @dondersinst.bsky.social is recruiting for two PhD and two postdoctoral positions funded by an @erc.europa.eu Consolidator - come join us!
PhD: www.mpi.nl/career-educa...
Postdoc: www.mpi.nl/career-educa...
(please share widely)
PhD: www.mpi.nl/career-educa...
Postdoc: www.mpi.nl/career-educa...
(please share widely)
Reposted by Vlad Niculae
Excited to share our ICLR 2025 oral "Residual Deep Gaussian Processes on Manifolds"!
With @vabor112.bsky.social & @arkrause.bsky.social, we introduce manifold-to-manifold GPs that can be composed together, generalising deep GPs to manifolds. Applications include wind prediction & Bayes opt! 1/n
With @vabor112.bsky.social & @arkrause.bsky.social, we introduce manifold-to-manifold GPs that can be composed together, generalising deep GPs to manifolds. Applications include wind prediction & Bayes opt! 1/n
February 13, 2025 at 4:45 PM
Excited to share our ICLR 2025 oral "Residual Deep Gaussian Processes on Manifolds"!
With @vabor112.bsky.social & @arkrause.bsky.social, we introduce manifold-to-manifold GPs that can be composed together, generalising deep GPs to manifolds. Applications include wind prediction & Bayes opt! 1/n
With @vabor112.bsky.social & @arkrause.bsky.social, we introduce manifold-to-manifold GPs that can be composed together, generalising deep GPs to manifolds. Applications include wind prediction & Bayes opt! 1/n
Reposted by Vlad Niculae
i can't believe how long we've spent fooling ourselves about the value of fully specified, massive matmuls instead of embracing the gods of sparsity
January 4, 2025 at 4:01 AM
i can't believe how long we've spent fooling ourselves about the value of fully specified, massive matmuls instead of embracing the gods of sparsity
Recruiting a PhD candidate at U. of Amsterdam (funded, 4yr). We will use ML&NLP, prob. models, and user studies, to make adaptive scientific-assistant systems that communicate & justify decisions in ways helpful to experts.
More: vene.ro/jobs.html
Apply by May 18: werkenbij.uva.nl/en/vacancies...
More: vene.ro/jobs.html
Apply by May 18: werkenbij.uva.nl/en/vacancies...
Vlad Niculae
vene.ro
April 24, 2025 at 1:03 PM
Recruiting a PhD candidate at U. of Amsterdam (funded, 4yr). We will use ML&NLP, prob. models, and user studies, to make adaptive scientific-assistant systems that communicate & justify decisions in ways helpful to experts.
More: vene.ro/jobs.html
Apply by May 18: werkenbij.uva.nl/en/vacancies...
More: vene.ro/jobs.html
Apply by May 18: werkenbij.uva.nl/en/vacancies...
Reposted by Vlad Niculae
Variational approximation with Gaussian mixtures is looking cute! So here it's just gradient descent on K(q||p) for optimising the mixtures means & covariances & weights...
@lacerbi.bsky.social
@lacerbi.bsky.social
November 20, 2024 at 6:23 PM
Variational approximation with Gaussian mixtures is looking cute! So here it's just gradient descent on K(q||p) for optimising the mixtures means & covariances & weights...
@lacerbi.bsky.social
@lacerbi.bsky.social
Reposted by Vlad Niculae
This review paper by @guillaume-garrigos.com on SGD-related algorithms is a fantastic resource, offering elegant, self-contained, and concise proofs in a single, accessible reference. arxiv.org/pdf/2301.11235
January 29, 2025 at 4:15 PM
This review paper by @guillaume-garrigos.com on SGD-related algorithms is a fantastic resource, offering elegant, self-contained, and concise proofs in a single, accessible reference. arxiv.org/pdf/2301.11235
Reposted by Vlad Niculae
These phenomenon have been observed since early vision systems. It is important to report these things, though. Maybe it will permeate and we won’t keep making the same mistakes over and over
They trained an AI model on a widely used knee osteoarthritis dataset to see if it would be able to make nonsensical predictions - whether the patient ate refried beans, or drank beer. It did, in part by somehow figuring out where the x-ray was taken.
www.dartmouth-health.org/about/news/a...
www.dartmouth-health.org/about/news/a...
AI thought X-rays of your knees show if you drink beer—they don’t.
Dartmouth Health study shows how easily AI models can give right answers for wrong reasons
www.dartmouth-health.org
December 14, 2024 at 6:25 PM
These phenomenon have been observed since early vision systems. It is important to report these things, though. Maybe it will permeate and we won’t keep making the same mistakes over and over
Reposted by Vlad Niculae
This is such a beautiful algorithm (and a nice analysis): to check if an array is sorted vs. far from being sorted (many entries need to be changed), just:
- pick an element uniformly at random in the array
- "forget" where it was
- try to find it again via binary search
Repeat this a few times.
- pick an element uniformly at random in the array
- "forget" where it was
- try to find it again via binary search
Repeat this a few times.
<spoiler>
The analysis is not obvious, but yes, that's the idea!
* Repeat O(1) times:
- Pick an index i uniformly at random, let x←A[i]
- Do a binary search for x in A, end at index j
- Return UNSORTED if i≠j
* Return SORTED
</spoiler>
The analysis is not obvious, but yes, that's the idea!
* Repeat O(1) times:
- Pick an index i uniformly at random, let x←A[i]
- Do a binary search for x in A, end at index j
- Return UNSORTED if i≠j
* Return SORTED
</spoiler>
December 14, 2024 at 10:50 AM
This is such a beautiful algorithm (and a nice analysis): to check if an array is sorted vs. far from being sorted (many entries need to be changed), just:
- pick an element uniformly at random in the array
- "forget" where it was
- try to find it again via binary search
Repeat this a few times.
- pick an element uniformly at random in the array
- "forget" where it was
- try to find it again via binary search
Repeat this a few times.
I and hundreds other workers at the University of Amsterdam are on strike with @fnv.bsky.social
www.linkedin.com/pulse/our-ha...
www.linkedin.com/pulse/our-ha...
It is in our hands: To protect our safety and the right to protest, do more than re-formulating the house rules.
Column by anonymous FNV-member from UvA Protest is a fundamental right, and universities have a duty to facilitate and protect it. The violent events surrounding pro-Palestine protests this past year ...
www.linkedin.com
December 12, 2024 at 12:08 PM
I and hundreds other workers at the University of Amsterdam are on strike with @fnv.bsky.social
www.linkedin.com/pulse/our-ha...
www.linkedin.com/pulse/our-ha...
Reposted by Vlad Niculae
"AI can be bad but also it can be good" is just a really dumb way to talk about anything...it's the grade-school exercise of "make a list of pros and cons" but pressed into service for producing a sense of inevitability and making the medicine go down
December 6, 2024 at 5:17 AM
"AI can be bad but also it can be good" is just a really dumb way to talk about anything...it's the grade-school exercise of "make a list of pros and cons" but pressed into service for producing a sense of inevitability and making the medicine go down
Reposted by Vlad Niculae
OpenAI in 2024:
“No AI for weapons or military”
“Do use our AI to make weapons to hurt yourself or others”
“Military is fine, but no AI for weapons”
“Sure put it on battlefield drones”
www.technologyreview.com/2024/12/04/1...
“No AI for weapons or military”
“Do use our AI to make weapons to hurt yourself or others”
“Military is fine, but no AI for weapons”
“Sure put it on battlefield drones”
www.technologyreview.com/2024/12/04/1...
OpenAI’s new defense contract completes its military pivot
A new partnership with Anduril, announced today, will deploy AI on the battlefield. It represents an overhaul of the company’s position in just a year.
www.technologyreview.com
December 4, 2024 at 11:43 PM
OpenAI in 2024:
“No AI for weapons or military”
“Do use our AI to make weapons to hurt yourself or others”
“Military is fine, but no AI for weapons”
“Sure put it on battlefield drones”
www.technologyreview.com/2024/12/04/1...
“No AI for weapons or military”
“Do use our AI to make weapons to hurt yourself or others”
“Military is fine, but no AI for weapons”
“Sure put it on battlefield drones”
www.technologyreview.com/2024/12/04/1...
Reposted by Vlad Niculae
Blue skies 🦋 , hot (?) takes 🔥
Constrained output for LLMs, e.g., outlines library for vllm which forces models to output json/pydantic schemas, is cool!
But, because output tokens cost much more latency than input tokens, if speed matters: bespoke, low-token output formats are often better.
Constrained output for LLMs, e.g., outlines library for vllm which forces models to output json/pydantic schemas, is cool!
But, because output tokens cost much more latency than input tokens, if speed matters: bespoke, low-token output formats are often better.
December 3, 2024 at 10:25 PM
Blue skies 🦋 , hot (?) takes 🔥
Constrained output for LLMs, e.g., outlines library for vllm which forces models to output json/pydantic schemas, is cool!
But, because output tokens cost much more latency than input tokens, if speed matters: bespoke, low-token output formats are often better.
Constrained output for LLMs, e.g., outlines library for vllm which forces models to output json/pydantic schemas, is cool!
But, because output tokens cost much more latency than input tokens, if speed matters: bespoke, low-token output formats are often better.
Reposted by Vlad Niculae
I hope I am not late to the party (was away post-quals chilling) but here are some thoughts on why this is bad IMO:
First, a disclaimer that I am writing this as an African who is a speaker of multiple African languages, NLP researcher of African languages, and HCI researcher focusing broadly on..
First, a disclaimer that I am writing this as an African who is a speaker of multiple African languages, NLP researcher of African languages, and HCI researcher focusing broadly on..
this is a green flag for openai & meta to formally be arbitrators of our languages & mass exploit the population (& researcher that've poured their souls into low resource languages),all to throw unreliable AI that has so far proven to result in more harm than benefit
www.reuters.com/technology/a...
www.reuters.com/technology/a...
Orange enlists Meta and OpenAI to develop AI language models in Africa
Orange will enlist OpenAI and Meta to fine-tune AI large language models (LLMs) to translate regional African languages for the French telecoms operator, it said on Tuesday.
www.reuters.com
December 2, 2024 at 11:43 PM
I hope I am not late to the party (was away post-quals chilling) but here are some thoughts on why this is bad IMO:
First, a disclaimer that I am writing this as an African who is a speaker of multiple African languages, NLP researcher of African languages, and HCI researcher focusing broadly on..
First, a disclaimer that I am writing this as an African who is a speaker of multiple African languages, NLP researcher of African languages, and HCI researcher focusing broadly on..