Thomas F. Varley
@thosvarley.bsky.social
Dual PhD: Complex Systems & Computational Neuroscience - Postdoc at UVM in the Vermont Complex Systems Institute.
Information theory, synergy, and emergence.
Connoisseur of collapse phenomena
Information theory, synergy, and emergence.
Connoisseur of collapse phenomena
My dog sunk every last point into charisma
But it works for her (she got a piece of fried pickle moments after this photo was taken).
But it works for her (she got a piece of fried pickle moments after this photo was taken).
June 30, 2025 at 12:43 AM
My dog sunk every last point into charisma
But it works for her (she got a piece of fried pickle moments after this photo was taken).
But it works for her (she got a piece of fried pickle moments after this photo was taken).
Just passed the 1000 citations mark! Not bad for a second-year postdoc!
June 27, 2025 at 4:10 PM
Just passed the 1000 citations mark! Not bad for a second-year postdoc!
This started out as my own personal set of Neovim configs, but at a certain point, I realized that this could be of more general interest.
Everything works on Debian Linux, I'm confident it will work on other Linux distros, and it's probably good on OSX. If you use Windows, you're on your own.
Everything works on Debian Linux, I'm confident it will work on other Linux distros, and it's probably good on OSX. If you use Windows, you're on your own.
April 30, 2025 at 11:23 AM
This started out as my own personal set of Neovim configs, but at a certain point, I realized that this could be of more general interest.
Everything works on Debian Linux, I'm confident it will work on other Linux distros, and it's probably good on OSX. If you use Windows, you're on your own.
Everything works on Debian Linux, I'm confident it will work on other Linux distros, and it's probably good on OSX. If you use Windows, you're on your own.
Mentally I'm here
images.squarespace-cdn.com/content/v1/5...
images.squarespace-cdn.com/content/v1/5...
April 29, 2025 at 12:38 AM
Mentally I'm here
images.squarespace-cdn.com/content/v1/5...
images.squarespace-cdn.com/content/v1/5...
Finally, we asked about how dimensionality reduction algorithms like PCA (very popular among "neural manifold" researchers) alter higher-order information.
It turns out that, like FC networks, PCA is largely insensitive to synergy and prefers redundancy.
11/n
It turns out that, like FC networks, PCA is largely insensitive to synergy and prefers redundancy.
11/n
April 15, 2025 at 11:48 AM
Finally, we asked about how dimensionality reduction algorithms like PCA (very popular among "neural manifold" researchers) alter higher-order information.
It turns out that, like FC networks, PCA is largely insensitive to synergy and prefers redundancy.
11/n
It turns out that, like FC networks, PCA is largely insensitive to synergy and prefers redundancy.
11/n
We found that both the average "lifetime" of 3D dimensional cavities in brain data, and the number of such cavities was significantly correlated with the O-information.
This effect survived (albeit weaker) after rotating the fMRI data w/ PCA. 10/n
This effect survived (albeit weaker) after rotating the fMRI data w/ PCA. 10/n
April 15, 2025 at 11:48 AM
We found that both the average "lifetime" of 3D dimensional cavities in brain data, and the number of such cavities was significantly correlated with the O-information.
This effect survived (albeit weaker) after rotating the fMRI data w/ PCA. 10/n
This effect survived (albeit weaker) after rotating the fMRI data w/ PCA. 10/n
We also looked at naturalistic data: fMRI signals from a human brain (very popular for both TDA and information-theoretic analyses). 9/n
April 15, 2025 at 11:48 AM
We also looked at naturalistic data: fMRI signals from a human brain (very popular for both TDA and information-theoretic analyses). 9/n
Finally, are there any shapes that are intrinsically redundancy-dominated, rather than intrinsically synergy-dominated? Yes: knots. One-dimensional lines embedded in a 3D dimensional space always seem to be significantly redundant in a way the straight line isn't. 8/n
April 15, 2025 at 11:38 AM
Finally, are there any shapes that are intrinsically redundancy-dominated, rather than intrinsically synergy-dominated? Yes: knots. One-dimensional lines embedded in a 3D dimensional space always seem to be significantly redundant in a way the straight line isn't. 8/n
If you compute the O-information of a filled ball (rather than a hollow sphere), it vanishes. You can also get mixtures of "intrinsic" and "contextual" higher-order information. A hollow torus will always have synergy, but the exact amount depends on how it's embedded.
7/n
7/n
April 15, 2025 at 11:38 AM
If you compute the O-information of a filled ball (rather than a hollow sphere), it vanishes. You can also get mixtures of "intrinsic" and "contextual" higher-order information. A hollow torus will always have synergy, but the exact amount depends on how it's embedded.
7/n
7/n
In contrast, a hollow sphere is strongly synergistic, no matter how you rotate the point cloud (due to symmetry). So there is a sense in which the higher-order information in a 2D plane is "contextual", but the higher-order information in the sphere is "intrinsic".
6/n
6/n
April 15, 2025 at 11:38 AM
In contrast, a hollow sphere is strongly synergistic, no matter how you rotate the point cloud (due to symmetry). So there is a sense in which the higher-order information in a 2D plane is "contextual", but the higher-order information in the sphere is "intrinsic".
6/n
6/n
The first thing we realized is that, for some point clouds: orientation matters. A 1D line or 2D plane can be embedded in a 3D space so that it appears to have higher-order redundancy or synergy, but a PCA will rotate it in a way that erases all higher-order info.
5/n
5/n
April 15, 2025 at 11:38 AM
The first thing we realized is that, for some point clouds: orientation matters. A 1D line or 2D plane can be embedded in a 3D space so that it appears to have higher-order redundancy or synergy, but a PCA will rotate it in a way that erases all higher-order info.
5/n
5/n
Very intrigued by the word "safely" in PLoS CB's form letter saying they received my submission. Is the submission guarded by a dragon? Or do they go on some kind of vision quest to gates of the Editorial Manager and retrieve it?
April 14, 2025 at 2:13 PM
Very intrigued by the word "safely" in PLoS CB's form letter saying they received my submission. Is the submission guarded by a dragon? Or do they go on some kind of vision quest to gates of the Editorial Manager and retrieve it?
Querying my external brain: I'm trying to use back-propagation to optimize the behavior of the system, and the gradient is confusing to me. Ideally this curve should be monotonically increasing (the loss function is always the same). But it's clearly not.
Done in Taichi Lang w/ an ADAM optimizer.
Done in Taichi Lang w/ an ADAM optimizer.
April 1, 2025 at 11:32 PM
Querying my external brain: I'm trying to use back-propagation to optimize the behavior of the system, and the gradient is confusing to me. Ideally this curve should be monotonically increasing (the loss function is always the same). But it's clearly not.
Done in Taichi Lang w/ an ADAM optimizer.
Done in Taichi Lang w/ an ADAM optimizer.
I am unreasonably pleased with this - some screenshots of my Desktop with various default LMDE backgrounds.
March 21, 2025 at 9:30 PM
I am unreasonably pleased with this - some screenshots of my Desktop with various default LMDE backgrounds.
I give you: the most Reddit-tier Reddit post of all time.
March 11, 2025 at 1:21 PM
I give you: the most Reddit-tier Reddit post of all time.
Here's something very weird and extremely niche. If you do a partial entropy decomposition of the S-information, and a Kullback-Leibler decomposition of the TSE-complexity, they have almost exactly the same form, just reversed.
February 15, 2025 at 1:43 AM
Here's something very weird and extremely niche. If you do a partial entropy decomposition of the S-information, and a Kullback-Leibler decomposition of the TSE-complexity, they have almost exactly the same form, just reversed.
New blog post.
In 2025, the American right is basically 4chan circa 2014, while the American left has become Tumblr from the same era.
What a piece of Internet history can tell us about our current political hellscape.
synergies.substack.com/p/the-4chan-...
In 2025, the American right is basically 4chan circa 2014, while the American left has become Tumblr from the same era.
What a piece of Internet history can tell us about our current political hellscape.
synergies.substack.com/p/the-4chan-...
February 10, 2025 at 3:01 PM
New blog post.
In 2025, the American right is basically 4chan circa 2014, while the American left has become Tumblr from the same era.
What a piece of Internet history can tell us about our current political hellscape.
synergies.substack.com/p/the-4chan-...
In 2025, the American right is basically 4chan circa 2014, while the American left has become Tumblr from the same era.
What a piece of Internet history can tell us about our current political hellscape.
synergies.substack.com/p/the-4chan-...
Tbh, one thing I will say in favor of this round of President Trump is that liberals don't seem to be losing their minds in the same deranged way as they did back in 2017. Remember this stuff?
February 9, 2025 at 4:48 PM
Tbh, one thing I will say in favor of this round of President Trump is that liberals don't seem to be losing their minds in the same deranged way as they did back in 2017. Remember this stuff?
I hate it here.
February 9, 2025 at 2:07 AM
I hate it here.
Cursed Shannon entropy
February 6, 2025 at 5:58 PM
Cursed Shannon entropy
Wild that AI is even automating the jobs of the men in comments sections who reflexively post skepticism about sexual assault allegations.
Truly, the future is now.
Truly, the future is now.
January 14, 2025 at 6:40 PM
Wild that AI is even automating the jobs of the men in comments sections who reflexively post skepticism about sexual assault allegations.
Truly, the future is now.
Truly, the future is now.
My sister-in-law snapped what may be the greatest photo of my dog ever.
January 2, 2025 at 2:39 AM
My sister-in-law snapped what may be the greatest photo of my dog ever.
Apropos of nothing - here's my dog. I think she might be an alien.
November 20, 2024 at 10:14 PM
Apropos of nothing - here's my dog. I think she might be an alien.
Hello BlueSky! I'm a postdoc at the University of Vermont - I'm interested in multivariate information theory, higher-order interactions in complex systems, and the emergence (and collapse) of complexity.
I also enjoy juggling and glassblowing.
I also enjoy juggling and glassblowing.
October 10, 2023 at 1:06 AM
Hello BlueSky! I'm a postdoc at the University of Vermont - I'm interested in multivariate information theory, higher-order interactions in complex systems, and the emergence (and collapse) of complexity.
I also enjoy juggling and glassblowing.
I also enjoy juggling and glassblowing.