Mircea Petrache
@mirceasci.bsky.social
Doing maths research (geom. analysis, metric geometry, large point configs., opt. transport, etc) since ~2010, deep learning and applied research since ~2020.
https://sites.google.com/site/mircpetrache/home
https://sites.google.com/site/mircpetrache/home
Reposted by Mircea Petrache
How do brain areas control each other? 🧠🎛️
✨In our NeurIPS 2025 Spotlight paper, we introduce a data-driven framework to answer this question using deep learning, nonlinear control, and differential geometry.🧵⬇️
✨In our NeurIPS 2025 Spotlight paper, we introduce a data-driven framework to answer this question using deep learning, nonlinear control, and differential geometry.🧵⬇️
November 26, 2025 at 7:32 PM
How do brain areas control each other? 🧠🎛️
✨In our NeurIPS 2025 Spotlight paper, we introduce a data-driven framework to answer this question using deep learning, nonlinear control, and differential geometry.🧵⬇️
✨In our NeurIPS 2025 Spotlight paper, we introduce a data-driven framework to answer this question using deep learning, nonlinear control, and differential geometry.🧵⬇️
2 open faculty positions in Artificial Intelligence + Math/Stat/CS at UC Chile www.mathjobs.org/jobs/list/27...
November 15, 2025 at 1:32 AM
2 open faculty positions in Artificial Intelligence + Math/Stat/CS at UC Chile www.mathjobs.org/jobs/list/27...
Me: "suggest how to rebuttal a reviewer that clearly used ChatGPT to generate reviews"
Chatgpt: "you can quote to them the parts that are clearly generated by ChatGPT, explaining what the correct versions of the statements are [..]"
yeah.. it's called "Human Feedback or Perish"
Chatgpt: "you can quote to them the parts that are clearly generated by ChatGPT, explaining what the correct versions of the statements are [..]"
yeah.. it's called "Human Feedback or Perish"
November 13, 2025 at 12:40 PM
Me: "suggest how to rebuttal a reviewer that clearly used ChatGPT to generate reviews"
Chatgpt: "you can quote to them the parts that are clearly generated by ChatGPT, explaining what the correct versions of the statements are [..]"
yeah.. it's called "Human Feedback or Perish"
Chatgpt: "you can quote to them the parts that are clearly generated by ChatGPT, explaining what the correct versions of the statements are [..]"
yeah.. it's called "Human Feedback or Perish"
Reposted by Mircea Petrache
The Ponizsa illusion - Kanizsa triangle produces Ponzo illusion, but oddly it inverts apparent depth: the longer line looks closer! Also has an inversion effect like we studied in Altan et al 2025 Proc Roy Soc (doi.org/10.1098/rspb...) but it's also in reverse! 🤔
#visionscience #psychscisky
#visionscience #psychscisky
November 12, 2025 at 2:58 AM
The Ponizsa illusion - Kanizsa triangle produces Ponzo illusion, but oddly it inverts apparent depth: the longer line looks closer! Also has an inversion effect like we studied in Altan et al 2025 Proc Roy Soc (doi.org/10.1098/rspb...) but it's also in reverse! 🤔
#visionscience #psychscisky
#visionscience #psychscisky
My university's computer science department just opened a Human-Centered Computing position.. it'd be awesome to have more colleagues working in human-centered deep learning, so please apply! dcc.ing.uc.cl/departamento...
Vacantes académicas - Departamento de Ciencia de la Computación
Vacantes académicas Faculty Position in Human-Centered Information Systems The School of Engineering at the Pontificia Universidad Católica de Chile, one of the leading engineering academic institutio...
dcc.ing.uc.cl
October 22, 2025 at 5:57 PM
My university's computer science department just opened a Human-Centered Computing position.. it'd be awesome to have more colleagues working in human-centered deep learning, so please apply! dcc.ing.uc.cl/departamento...
I think this makes a lot of sense, although I'm still getting used to it
When I was a starting grad student (including masters), I used to love conference deadlines. They forced discipline, imposed a structure, and ensured packets of projects were finished in published form. Then sometime later, including during my postdocs, I developed disdain for conference
September 23, 2025 at 12:12 PM
I think this makes a lot of sense, although I'm still getting used to it
Reposted by Mircea Petrache
We are excited to share GPN-Star, a cost-effective, biologically grounded genomic language modeling framework that achieves state-of-the-art performance across a wide range of variant effect prediction tasks relevant to human genetics.
www.biorxiv.org/content/10.1...
(1/n)
www.biorxiv.org/content/10.1...
(1/n)
September 22, 2025 at 5:29 AM
We are excited to share GPN-Star, a cost-effective, biologically grounded genomic language modeling framework that achieves state-of-the-art performance across a wide range of variant effect prediction tasks relevant to human genetics.
www.biorxiv.org/content/10.1...
(1/n)
www.biorxiv.org/content/10.1...
(1/n)
2 tenure-track positions at UC Chile just opened.. in the maths/stats department. We have good growing links with the dept of computational engineering, so people wanting to fill the bridge maths-CS are welcome.
Have a look (and maybe click "apply") here mathjobs.org/jobs/list/27...
Have a look (and maybe click "apply") here mathjobs.org/jobs/list/27...
September 12, 2025 at 2:09 AM
2 tenure-track positions at UC Chile just opened.. in the maths/stats department. We have good growing links with the dept of computational engineering, so people wanting to fill the bridge maths-CS are welcome.
Have a look (and maybe click "apply") here mathjobs.org/jobs/list/27...
Have a look (and maybe click "apply") here mathjobs.org/jobs/list/27...
Reposted by Mircea Petrache
Nice paper arxiv.org/abs/2506.01622
General agents need world models
Are world models a necessary ingredient for flexible, goal-directed behaviour, or is model-free learning sufficient? We provide a formal answer to this question, showing that any agent capable of gene...
arxiv.org
July 9, 2025 at 1:57 PM
Nice paper arxiv.org/abs/2506.01622
Reposted by Mircea Petrache
Last minute twist on these positions: they are available for EVERYONE, not just EU nationals 🎉 Closing date is tomorrow on some adverts (I won't be able to find and fix them) but we will consider late applications through to early next week (Monday 16th) #roots #rootecolology #biogeochemistry #phd
June 10, 2025 at 3:47 PM
Last minute twist on these positions: they are available for EVERYONE, not just EU nationals 🎉 Closing date is tomorrow on some adverts (I won't be able to find and fix them) but we will consider late applications through to early next week (Monday 16th) #roots #rootecolology #biogeochemistry #phd
Fun fact: These C:N ratios are realistic values, but the mechanism is far more complex than "27=4+23"
Maths: (a+b)/(c+d)=a/c + b/d is never true for a,b,c,d>0.. first hint to dig deeper
Biology: main sink of C is respiration.. plant+hopper+frass/urea are NOT the only elements in the equation..
Maths: (a+b)/(c+d)=a/c + b/d is never true for a,b,c,d>0.. first hint to dig deeper
Biology: main sink of C is respiration.. plant+hopper+frass/urea are NOT the only elements in the equation..
Increased CO2 has the same effect. This is bad news for Earth's consumers, particularly plant eaters that already live on a nutritional knife edge and are trying to survive on less and less nutritious diet with predictable results: herbivore numbers are declining faster than their predators.
June 9, 2025 at 4:12 AM
Fun fact: These C:N ratios are realistic values, but the mechanism is far more complex than "27=4+23"
Maths: (a+b)/(c+d)=a/c + b/d is never true for a,b,c,d>0.. first hint to dig deeper
Biology: main sink of C is respiration.. plant+hopper+frass/urea are NOT the only elements in the equation..
Maths: (a+b)/(c+d)=a/c + b/d is never true for a,b,c,d>0.. first hint to dig deeper
Biology: main sink of C is respiration.. plant+hopper+frass/urea are NOT the only elements in the equation..
Reposted by Mircea Petrache
Nutrient Dilution and the Future of Herbivores: a short thread
The fossil fuel folks claimed that all the extra CO2 is great for plants. Indeed, plant mass has increased globally by about 30% in the past century.
But when you load a plant up the carbs, what happens to all its other nutrients?
The fossil fuel folks claimed that all the extra CO2 is great for plants. Indeed, plant mass has increased globally by about 30% in the past century.
But when you load a plant up the carbs, what happens to all its other nutrients?
November 18, 2024 at 9:31 PM
Nutrient Dilution and the Future of Herbivores: a short thread
The fossil fuel folks claimed that all the extra CO2 is great for plants. Indeed, plant mass has increased globally by about 30% in the past century.
But when you load a plant up the carbs, what happens to all its other nutrients?
The fossil fuel folks claimed that all the extra CO2 is great for plants. Indeed, plant mass has increased globally by about 30% in the past century.
But when you load a plant up the carbs, what happens to all its other nutrients?
Reposted by Mircea Petrache
Hot off the presses and so so thrilled. Thanks to @inaturalist.bsky.social + computer vision we found something beautifully simple: Red and orange flowers bloom later than all the other colors in the eastern United States. Paper here: authors.elsevier.com/a/1kwzh3QW8S...
April 14, 2025 at 9:50 PM
Hot off the presses and so so thrilled. Thanks to @inaturalist.bsky.social + computer vision we found something beautifully simple: Red and orange flowers bloom later than all the other colors in the eastern United States. Paper here: authors.elsevier.com/a/1kwzh3QW8S...
Reposted by Mircea Petrache
It's fascinating that generating accurate references is so difficult for LLMs, no matter how much better they become. You can clearly specify the source, instruct the model to search, and even watch it retrieve the correct paper--yet it may still produce a fictitious citation.
June 1, 2025 at 3:56 PM
It's fascinating that generating accurate references is so difficult for LLMs, no matter how much better they become. You can clearly specify the source, instruct the model to search, and even watch it retrieve the correct paper--yet it may still produce a fictitious citation.
Learned a cool word today: "tivrasamveg".. been doing lots of that recently
May 29, 2025 at 11:23 PM
Learned a cool word today: "tivrasamveg".. been doing lots of that recently
Reposted by Mircea Petrache
New preprint alert 🚨
How can you guide diffusion and flow-based generative models when data is scarce but you have domain knowledge? We introduce Minimum Excess Work, a physics-inspired method for efficiently integrating sparse constraints.
Thread below 👇https://arxiv.org/abs/2505.13375
How can you guide diffusion and flow-based generative models when data is scarce but you have domain knowledge? We introduce Minimum Excess Work, a physics-inspired method for efficiently integrating sparse constraints.
Thread below 👇https://arxiv.org/abs/2505.13375
May 26, 2025 at 9:13 AM
New preprint alert 🚨
How can you guide diffusion and flow-based generative models when data is scarce but you have domain knowledge? We introduce Minimum Excess Work, a physics-inspired method for efficiently integrating sparse constraints.
Thread below 👇https://arxiv.org/abs/2505.13375
How can you guide diffusion and flow-based generative models when data is scarce but you have domain knowledge? We introduce Minimum Excess Work, a physics-inspired method for efficiently integrating sparse constraints.
Thread below 👇https://arxiv.org/abs/2505.13375
Reposted by Mircea Petrache
1/6 Why does the brain maintain such precise excitatory-inhibitory balance?
Our new preprint explores a provocative idea: Small, targeted deviations from this balance may serve a purpose: to encode local error signals for learning.
www.biorxiv.org/content/10.1...
led by @jrbch.bsky.social
Our new preprint explores a provocative idea: Small, targeted deviations from this balance may serve a purpose: to encode local error signals for learning.
www.biorxiv.org/content/10.1...
led by @jrbch.bsky.social
May 27, 2025 at 7:49 AM
1/6 Why does the brain maintain such precise excitatory-inhibitory balance?
Our new preprint explores a provocative idea: Small, targeted deviations from this balance may serve a purpose: to encode local error signals for learning.
www.biorxiv.org/content/10.1...
led by @jrbch.bsky.social
Our new preprint explores a provocative idea: Small, targeted deviations from this balance may serve a purpose: to encode local error signals for learning.
www.biorxiv.org/content/10.1...
led by @jrbch.bsky.social
www.youtube.com/watch?v=MVGc... this is from 1849
Charles-Valentin Alkan - Étude WoO from Encyclopédie du Pianiste Compositeur (Viner)
YouTube video by The Alkan Society
www.youtube.com
May 19, 2025 at 1:12 AM
www.youtube.com/watch?v=MVGc... this is from 1849
Been studying dungeon synth for the past few weeks. It's a quite diverse and creative genre. If I had more time on my hands, creating stuff like this would be definitely in my top 5 hobby list.. for now listening and trying to chant them back to a 7mo baby is enough www.youtube.com/watch?v=iwGE...
Forgotten Pathways - Shrouded in Mystery (1998) (Dungeon Synth)
YouTube video by The Dungeon Synth Archives
www.youtube.com
May 17, 2025 at 1:03 AM
Been studying dungeon synth for the past few weeks. It's a quite diverse and creative genre. If I had more time on my hands, creating stuff like this would be definitely in my top 5 hobby list.. for now listening and trying to chant them back to a 7mo baby is enough www.youtube.com/watch?v=iwGE...
www.youtube.com/watch?v=NcqF... I didn't believe the video title at first.. so cool
Bizarre Tree That Evolved to Use Lightning for Survival
YouTube video by Anton Petrov
www.youtube.com
April 17, 2025 at 3:48 AM
www.youtube.com/watch?v=NcqF... I didn't believe the video title at first.. so cool
Reposted by Mircea Petrache
Super excited to share that the 𝗦𝗘𝗩𝗔 dataset and benchmark will be presented at #NeurIPS2023!✏️🧠🤖
🌐 Website - seva-benchmark.github.io
📜 Paper - arxiv.org/abs/2312.03035
💻 Code + Materials - github.com/cogtoolslab/...
Thread below 🧵-
🌐 Website - seva-benchmark.github.io
📜 Paper - arxiv.org/abs/2312.03035
💻 Code + Materials - github.com/cogtoolslab/...
Thread below 🧵-
December 8, 2023 at 2:26 AM
Super excited to share that the 𝗦𝗘𝗩𝗔 dataset and benchmark will be presented at #NeurIPS2023!✏️🧠🤖
🌐 Website - seva-benchmark.github.io
📜 Paper - arxiv.org/abs/2312.03035
💻 Code + Materials - github.com/cogtoolslab/...
Thread below 🧵-
🌐 Website - seva-benchmark.github.io
📜 Paper - arxiv.org/abs/2312.03035
💻 Code + Materials - github.com/cogtoolslab/...
Thread below 🧵-
Question for #neuroscience people:
On a population of neurons and I can do 2 possible stimulations with equal total intensity and duration:
(1) periodic stimulation, same intensity for all pulses
(2) random stimulation, random intensity
The question: will (1) or (2) elicit more spikes in total?
On a population of neurons and I can do 2 possible stimulations with equal total intensity and duration:
(1) periodic stimulation, same intensity for all pulses
(2) random stimulation, random intensity
The question: will (1) or (2) elicit more spikes in total?
March 29, 2025 at 3:51 PM
Question for #neuroscience people:
On a population of neurons and I can do 2 possible stimulations with equal total intensity and duration:
(1) periodic stimulation, same intensity for all pulses
(2) random stimulation, random intensity
The question: will (1) or (2) elicit more spikes in total?
On a population of neurons and I can do 2 possible stimulations with equal total intensity and duration:
(1) periodic stimulation, same intensity for all pulses
(2) random stimulation, random intensity
The question: will (1) or (2) elicit more spikes in total?
today I'm heavily falling prey to the "I write much better reviews than the ones that I receive" fallacy
March 25, 2025 at 1:17 PM
today I'm heavily falling prey to the "I write much better reviews than the ones that I receive" fallacy
Wow, that's remarkable
The Swiss Federal Office of Topography has released high-precision LiDAR data of Switzerland:
www.swisstopo.admin.ch/en/height-mo...
www.swisstopo.admin.ch/en/height-mo...
March 21, 2025 at 2:14 PM
Wow, that's remarkable