Andreas Grivas
@andreasgrv.bsky.social
Postdoc in ML/NLP at the University of Edinburgh.
Interested in Bottlenecks in Neural Networks; Unargmaxable Outputs.
https://grv.unargmaxable.ai/
Interested in Bottlenecks in Neural Networks; Unargmaxable Outputs.
https://grv.unargmaxable.ai/
Pinned
Andreas Grivas
@andreasgrv.bsky.social
· Jan 10
📕 ⬇️ My thesis on 🚫unargmaxable outputs is online! Check it out if you want to learn more about how output layers constrain what neural networks can and cannot predict 👉 era.ed.ac.uk/handle/1842/...
Reposted by Andreas Grivas
We will be running the 13th iteration of the workshop on Fine-Grained Visual Categorization (FGVC) at CVPR this summer. We have an open call for papers that explore questions in visual recognition.
Please see our website for more information:
sites.google.com/view/fgvc13/...
Please see our website for more information:
sites.google.com/view/fgvc13/...
FGVC's not dead!
The 13th Workshop on Fine-Grained Visual Categorization has been accepted to CVPR 2026, in Denver, Colorado!
CALL FOR PAPERS: sites.google.com/view/fgvc13/
From Ecology to Medical Imagining, join us as we tackle the long tail and the limits of visual discrimination! #CVPR2026 #AI
The 13th Workshop on Fine-Grained Visual Categorization has been accepted to CVPR 2026, in Denver, Colorado!
CALL FOR PAPERS: sites.google.com/view/fgvc13/
From Ecology to Medical Imagining, join us as we tackle the long tail and the limits of visual discrimination! #CVPR2026 #AI
January 13, 2026 at 6:26 PM
We will be running the 13th iteration of the workshop on Fine-Grained Visual Categorization (FGVC) at CVPR this summer. We have an open call for papers that explore questions in visual recognition.
Please see our website for more information:
sites.google.com/view/fgvc13/...
Please see our website for more information:
sites.google.com/view/fgvc13/...
Fantastic opportunity to work on #reliable #LLMs in Edinburgh! Antonio is phenomenal and super fun to work with!
🚨🚨🚨
I have a #postdoc (1+1 yrs) at the School of Informatics, The University of Edinburgh to work on 💥 #reliable #LLM #agents with #neurosymbolic #nesy layers 💥 with Huawei Trustworthy Technology and Engineering Laboratory Munich
👉 elxw.fa.em3.oraclecloud.com/hcmUI/Candid...
🙏 please share!
I have a #postdoc (1+1 yrs) at the School of Informatics, The University of Edinburgh to work on 💥 #reliable #LLM #agents with #neurosymbolic #nesy layers 💥 with Huawei Trustworthy Technology and Engineering Laboratory Munich
👉 elxw.fa.em3.oraclecloud.com/hcmUI/Candid...
🙏 please share!
Research Associate
The School of Informatics, University of Edinburgh invites applications for a 1-year Post-Doctoral Research Associate (PDRA), to do research on reliable large language agents via neuro-symbolic integr...
elxw.fa.em3.oraclecloud.com
December 24, 2025 at 3:08 PM
Yeah? Well, you know, that's just like your slopinion, man. (2025 version)
a man in a green shirt sits with his hands on his head in a bowling alley
Alt: Lebowski from the film "The Big Lebowski" saying: yeah? well, you know, that's just like your opinion, man.
media.tenor.com
December 22, 2025 at 1:16 PM
Yeah? Well, you know, that's just like your slopinion, man. (2025 version)
Reposted by Andreas Grivas
Introducing Bolmo, a new family of byte-level language models built by "byteifying" our open Olmo 3—and to our knowledge, the first fully open byte-level LM to match or surpass SOTA subword models across a wide range of tasks. 🧵
December 15, 2025 at 5:19 PM
Introducing Bolmo, a new family of byte-level language models built by "byteifying" our open Olmo 3—and to our knowledge, the first fully open byte-level LM to match or surpass SOTA subword models across a wide range of tasks. 🧵
Reposted by Andreas Grivas
📣 PhD opening – Fall 2026
The DUCK Lab @imperialcollegeldn.bsky.social is looking for a PhD student to join us!
Why 🦆? We work on foundational aspects of #neurosymbolicAI and #SafeAI.
👉 DUCK = Data, Uncertainty, Constraints & Knowledge
📩 Apply by emailing: [email protected]
The DUCK Lab @imperialcollegeldn.bsky.social is looking for a PhD student to join us!
Why 🦆? We work on foundational aspects of #neurosymbolicAI and #SafeAI.
👉 DUCK = Data, Uncertainty, Constraints & Knowledge
📩 Apply by emailing: [email protected]
December 15, 2025 at 11:47 AM
📣 PhD opening – Fall 2026
The DUCK Lab @imperialcollegeldn.bsky.social is looking for a PhD student to join us!
Why 🦆? We work on foundational aspects of #neurosymbolicAI and #SafeAI.
👉 DUCK = Data, Uncertainty, Constraints & Knowledge
📩 Apply by emailing: [email protected]
The DUCK Lab @imperialcollegeldn.bsky.social is looking for a PhD student to join us!
Why 🦆? We work on foundational aspects of #neurosymbolicAI and #SafeAI.
👉 DUCK = Data, Uncertainty, Constraints & Knowledge
📩 Apply by emailing: [email protected]
Reposted by Andreas Grivas
I will be presenting this work from 11AM-2PM at #NeurIPS2025 in San Diego today! Come by poster #2012 to learn more :)
From medicine to geo-guessing, humans can get incredibly good at solving visual recognition tasks.
But how is this skill learned, and can we model its progression?
We present CleverBirds, accepted #NeurIPS2025, a large-scale benchmark for visual knowledge tracing.
📄 arxiv.org/abs/2511.08512
1/5
But how is this skill learned, and can we model its progression?
We present CleverBirds, accepted #NeurIPS2025, a large-scale benchmark for visual knowledge tracing.
📄 arxiv.org/abs/2511.08512
1/5
CleverBirds: A Multiple-Choice Benchmark for Fine-grained Human Knowledge Tracing
Mastering fine-grained visual recognition, essential in many expert domains, can require that specialists undergo years of dedicated training. Modeling the progression of such expertize in humans rema...
arxiv.org
December 4, 2025 at 5:15 PM
I will be presenting this work from 11AM-2PM at #NeurIPS2025 in San Diego today! Come by poster #2012 to learn more :)
Reposted by Andreas Grivas
At #EMNLP2025 to present the last chapter of my PhD 🐼
Let's talk #HateSpeech detection, generalisation and NLP safety at my poster:
📆tomorrow
🕟4.30pm
Look for the circus-themed poster 🎪🤸🏻♀️
Work with @tomsherborne.bsky.social @bjoernross.bsky.social and Mirella Lapata at #EdinburghNLP + @cohere.com
Let's talk #HateSpeech detection, generalisation and NLP safety at my poster:
📆tomorrow
🕟4.30pm
Look for the circus-themed poster 🎪🤸🏻♀️
Work with @tomsherborne.bsky.social @bjoernross.bsky.social and Mirella Lapata at #EdinburghNLP + @cohere.com
November 4, 2025 at 1:01 PM
At #EMNLP2025 to present the last chapter of my PhD 🐼
Let's talk #HateSpeech detection, generalisation and NLP safety at my poster:
📆tomorrow
🕟4.30pm
Look for the circus-themed poster 🎪🤸🏻♀️
Work with @tomsherborne.bsky.social @bjoernross.bsky.social and Mirella Lapata at #EdinburghNLP + @cohere.com
Let's talk #HateSpeech detection, generalisation and NLP safety at my poster:
📆tomorrow
🕟4.30pm
Look for the circus-themed poster 🎪🤸🏻♀️
Work with @tomsherborne.bsky.social @bjoernross.bsky.social and Mirella Lapata at #EdinburghNLP + @cohere.com
Reposted by Andreas Grivas
Reposted by Andreas Grivas
Reposted by Andreas Grivas
Edinburgh castle failing to render, likely thanks to the AWS outage
October 20, 2025 at 8:35 AM
Edinburgh castle failing to render, likely thanks to the AWS outage
Just submitted an application, picked up a banana and was greeted by:
October 20, 2025 at 2:51 PM
Just submitted an application, picked up a banana and was greeted by:
Reposted by Andreas Grivas
Interested in doing a PhD in machine learning at the University of Edinburgh starting Sept 2026?
My group works on topics in vision, machine learning, and AI for climate.
For more information and details on how to get in touch, please check out my website:
homepages.inf.ed.ac.uk/omacaod
My group works on topics in vision, machine learning, and AI for climate.
For more information and details on how to get in touch, please check out my website:
homepages.inf.ed.ac.uk/omacaod
October 16, 2025 at 9:15 AM
Interested in doing a PhD in machine learning at the University of Edinburgh starting Sept 2026?
My group works on topics in vision, machine learning, and AI for climate.
For more information and details on how to get in touch, please check out my website:
homepages.inf.ed.ac.uk/omacaod
My group works on topics in vision, machine learning, and AI for climate.
For more information and details on how to get in touch, please check out my website:
homepages.inf.ed.ac.uk/omacaod
LLM log probs approx lie on low dimensional ellipses which are hard to "forge", check out Matt's cool work :)
We discovered that language models leave a natural "signature" on their API outputs that's extremely hard to fake. Here's how it works 🔍
📄 arxiv.org/abs/2510.14086 1/
📄 arxiv.org/abs/2510.14086 1/
Every Language Model Has a Forgery-Resistant Signature
The ubiquity of closed-weight language models with public-facing APIs has generated interest in forensic methods, both for extracting hidden model details (e.g., parameters) and for identifying...
arxiv.org
October 17, 2025 at 6:23 PM
LLM log probs approx lie on low dimensional ellipses which are hard to "forge", check out Matt's cool work :)
Reposted by Andreas Grivas
We’re excited to see @nsaphra.bsky.social’s work featured in @quantamagazine.bsky.social—& even more excited that she will be joining the CDS faculty in 2026!
Read about Naomi's research via @benbenbrubaker.bsky.social spr.ly/63323AbqfH
📸 Ken Richardson
Read about Naomi's research via @benbenbrubaker.bsky.social spr.ly/63323AbqfH
📸 Ken Richardson
September 25, 2025 at 12:00 PM
We’re excited to see @nsaphra.bsky.social’s work featured in @quantamagazine.bsky.social—& even more excited that she will be joining the CDS faculty in 2026!
Read about Naomi's research via @benbenbrubaker.bsky.social spr.ly/63323AbqfH
📸 Ken Richardson
Read about Naomi's research via @benbenbrubaker.bsky.social spr.ly/63323AbqfH
📸 Ken Richardson
The best feature of overleaf is the subliminal messages it produces by choosing how to order the author tiles.
It was a long day, and Overleaf was like, maybe you should go to the pub!
It was a long day, and Overleaf was like, maybe you should go to the pub!
September 25, 2025 at 12:57 PM
The best feature of overleaf is the subliminal messages it produces by choosing how to order the author tiles.
It was a long day, and Overleaf was like, maybe you should go to the pub!
It was a long day, and Overleaf was like, maybe you should go to the pub!
Reposted by Andreas Grivas
It's been a while, but I am happy to share that my PhD dissertation is finally available online! 🎉
Not only it contains most of my work, but there is plenty of brand new content:
publikationen.sulb.uni-saarland.de/handle/20.50...
🧵1/4
Not only it contains most of my work, but there is plenty of brand new content:
publikationen.sulb.uni-saarland.de/handle/20.50...
🧵1/4
Publikationen der UdS: Meet my expectations: on the interplay of trustworthiness and deep learning optimization
publikationen.sulb.uni-saarland.de
August 6, 2025 at 7:57 PM
It's been a while, but I am happy to share that my PhD dissertation is finally available online! 🎉
Not only it contains most of my work, but there is plenty of brand new content:
publikationen.sulb.uni-saarland.de/handle/20.50...
🧵1/4
Not only it contains most of my work, but there is plenty of brand new content:
publikationen.sulb.uni-saarland.de/handle/20.50...
🧵1/4
Reposted by Andreas Grivas
At #ACL2025NLP and on the job market (NLP + AI Safety) 💼
It's great to see growing interest in safety/alignment, but we often miss the social context.
Come to our @woahworkshop.bsky.social Friday to dive deeper into safe safety research!
A quiet token from the biggest @aclmeeting.bsky.social ⬇️
It's great to see growing interest in safety/alignment, but we often miss the social context.
Come to our @woahworkshop.bsky.social Friday to dive deeper into safe safety research!
A quiet token from the biggest @aclmeeting.bsky.social ⬇️
July 29, 2025 at 9:54 AM
At #ACL2025NLP and on the job market (NLP + AI Safety) 💼
It's great to see growing interest in safety/alignment, but we often miss the social context.
Come to our @woahworkshop.bsky.social Friday to dive deeper into safe safety research!
A quiet token from the biggest @aclmeeting.bsky.social ⬇️
It's great to see growing interest in safety/alignment, but we often miss the social context.
Come to our @woahworkshop.bsky.social Friday to dive deeper into safe safety research!
A quiet token from the biggest @aclmeeting.bsky.social ⬇️
Reposted by Andreas Grivas
Next week I will be in 🌴 Rio 🌴 attending (and helping organizing) UAI 2025 and TPM 2025.
If you are also attending and want to talk science (e.g. probabilistic generative models, causality, multi-objective machine learning, etc) or just hang out, not everything is work, feel free to reach me out!
If you are also attending and want to talk science (e.g. probabilistic generative models, causality, multi-objective machine learning, etc) or just hang out, not everything is work, feel free to reach me out!
This year I am co-organizing the 8th iteration of the Tractable Probabilistic Modeling #TPM workshop at #UAI2025 🌴 Rio de Janeiro edition 🌴
🌐 lnkd.in/dDK8T5Au
⏰ Submission deadline: May 23th AoE
🌴 Date: July 15th
🧵👇
🌐 lnkd.in/dDK8T5Au
⏰ Submission deadline: May 23th AoE
🌴 Date: July 15th
🧵👇
July 18, 2025 at 11:17 AM
Next week I will be in 🌴 Rio 🌴 attending (and helping organizing) UAI 2025 and TPM 2025.
If you are also attending and want to talk science (e.g. probabilistic generative models, causality, multi-objective machine learning, etc) or just hang out, not everything is work, feel free to reach me out!
If you are also attending and want to talk science (e.g. probabilistic generative models, causality, multi-objective machine learning, etc) or just hang out, not everything is work, feel free to reach me out!
Reposted by Andreas Grivas
by applying a QR decomposition, any matrix can be written as a QR code multiplied by an upper triangular matrix
July 14, 2025 at 4:06 PM
by applying a QR decomposition, any matrix can be written as a QR code multiplied by an upper triangular matrix
Reposted by Andreas Grivas
📢I am hiring a Postdoc to work on post-training methods for low-resource languages. Apply by August 15 employment.ku.dk/faculty/?sho....
Let's talk at #ACL2025NLP in Vienna if you want to know more about the position and life in Denmark.
Let's talk at #ACL2025NLP in Vienna if you want to know more about the position and life in Denmark.
Postdoc in Natural Language Processing
employment.ku.dk
July 7, 2025 at 12:47 PM
📢I am hiring a Postdoc to work on post-training methods for low-resource languages. Apply by August 15 employment.ku.dk/faculty/?sho....
Let's talk at #ACL2025NLP in Vienna if you want to know more about the position and life in Denmark.
Let's talk at #ACL2025NLP in Vienna if you want to know more about the position and life in Denmark.
Reposted by Andreas Grivas
"in 2025 we will have flying cars" 😂😂😂
July 5, 2025 at 4:17 PM
"in 2025 we will have flying cars" 😂😂😂
Reposted by Andreas Grivas
this profile is impersonating me on X and scamming people. And it got verified...! (*)
golden rule: it is not me on social media if you don't see a ☠️.
(*) please help me report this profile, if you are still on X
golden rule: it is not me on social media if you don't see a ☠️.
(*) please help me report this profile, if you are still on X
June 25, 2025 at 6:05 AM
this profile is impersonating me on X and scamming people. And it got verified...! (*)
golden rule: it is not me on social media if you don't see a ☠️.
(*) please help me report this profile, if you are still on X
golden rule: it is not me on social media if you don't see a ☠️.
(*) please help me report this profile, if you are still on X
Reposted by Andreas Grivas
Dear #mathsky, if you repost this, it will let people see the article for free. Thanks!
www.nytimes.com/interactive/...
www.nytimes.com/interactive/...
How Bees, Beer Cans and Data Solve the Same Packing Problem (Gift Article)
Trying to fit it all in? There’s a trick to it, even in 24 dimensions.
www.nytimes.com
June 23, 2025 at 10:53 AM
Dear #mathsky, if you repost this, it will let people see the article for free. Thanks!
www.nytimes.com/interactive/...
www.nytimes.com/interactive/...
Reposted by Andreas Grivas
☀️ I will be in Edinburgh this summer (Jun-Aug) at @edinburgh-uni.bsky.social, visiting the group of Antonio Vergari (@nolovedeeplearning.bsky.social) to work on #neurosymbolic #AI and #probabilistic #ML
If you are around and want to chat about these topics, let's meet for coffee!
If you are around and want to chat about these topics, let's meet for coffee!
June 12, 2025 at 10:08 AM
☀️ I will be in Edinburgh this summer (Jun-Aug) at @edinburgh-uni.bsky.social, visiting the group of Antonio Vergari (@nolovedeeplearning.bsky.social) to work on #neurosymbolic #AI and #probabilistic #ML
If you are around and want to chat about these topics, let's meet for coffee!
If you are around and want to chat about these topics, let's meet for coffee!
Reposted by Andreas Grivas
🚀 By *learning* to compress the KV cache in Transformer LLMs, we can generate more tokens for the same compute budget.
This unlocks *inference-time hyper-scaling*
For the same runtime or memory load, we can boost LLM accuracy by pushing reasoning even further!
This unlocks *inference-time hyper-scaling*
For the same runtime or memory load, we can boost LLM accuracy by pushing reasoning even further!
June 6, 2025 at 12:33 PM
🚀 By *learning* to compress the KV cache in Transformer LLMs, we can generate more tokens for the same compute budget.
This unlocks *inference-time hyper-scaling*
For the same runtime or memory load, we can boost LLM accuracy by pushing reasoning even further!
This unlocks *inference-time hyper-scaling*
For the same runtime or memory load, we can boost LLM accuracy by pushing reasoning even further!