Alex Williams
@itsneuronal.bsky.social
Asst Prof at NYU + Flatiron Institute
Computational + Statistical Neuroscience
https://neurostatslab.org/
Computational + Statistical Neuroscience
https://neurostatslab.org/
Oh and they said "death to America", which is a kind of saber-rattling that is unthinkable to us at the New York Times.
June 22, 2025 at 4:08 PM
Oh and they said "death to America", which is a kind of saber-rattling that is unthinkable to us at the New York Times.
Opening sentence: "Iran has a long history of animosity towards the US"
Supporting evidence: "When we assassinated their top general without an open declaration of war they responded by token airstrikes against military bases that we installed at their doorstep in violation of international law"
Supporting evidence: "When we assassinated their top general without an open declaration of war they responded by token airstrikes against military bases that we installed at their doorstep in violation of international law"
June 22, 2025 at 4:08 PM
Opening sentence: "Iran has a long history of animosity towards the US"
Supporting evidence: "When we assassinated their top general without an open declaration of war they responded by token airstrikes against military bases that we installed at their doorstep in violation of international law"
Supporting evidence: "When we assassinated their top general without an open declaration of war they responded by token airstrikes against military bases that we installed at their doorstep in violation of international law"
Columbia leadership released another statement doubling down on the line that the government's concerns are "legitimate".
Another astounding abdication of responsibility.
Another astounding abdication of responsibility.
March 10, 2025 at 10:37 PM
Columbia leadership released another statement doubling down on the line that the government's concerns are "legitimate".
Another astounding abdication of responsibility.
Another astounding abdication of responsibility.
I am gobsmacked that this is the response Columbia leadership went with.
"Legitimate concerns"?!?! Are you kidding me?!?!
This action is illegal. It is being carried out on a transparently fake pretenses. People livelihoods are at risk! WTF are you doing?!?!
president.columbia.edu/news/respond...
"Legitimate concerns"?!?! Are you kidding me?!?!
This action is illegal. It is being carried out on a transparently fake pretenses. People livelihoods are at risk! WTF are you doing?!?!
president.columbia.edu/news/respond...
March 7, 2025 at 11:25 PM
I am gobsmacked that this is the response Columbia leadership went with.
"Legitimate concerns"?!?! Are you kidding me?!?!
This action is illegal. It is being carried out on a transparently fake pretenses. People livelihoods are at risk! WTF are you doing?!?!
president.columbia.edu/news/respond...
"Legitimate concerns"?!?! Are you kidding me?!?!
This action is illegal. It is being carried out on a transparently fake pretenses. People livelihoods are at risk! WTF are you doing?!?!
president.columbia.edu/news/respond...
Eve Marder was doing this before* you were born!
(*exaggerated for effect)
(*exaggerated for effect)
November 19, 2024 at 1:21 PM
Eve Marder was doing this before* you were born!
(*exaggerated for effect)
(*exaggerated for effect)
We also show that Procrustes shape distance (an alternative to CKA/RSA) is related to decoding but is generally more stringent.
The figure below shows a regime (panel D) where the Procrustes distance (x-axis) can be quite large even when the distance in terms of decoding is small (y-axis).
The figure below shows a regime (panel D) where the Procrustes distance (x-axis) can be quite large even when the distance in terms of decoding is small (y-axis).
November 18, 2024 at 6:18 PM
We also show that Procrustes shape distance (an alternative to CKA/RSA) is related to decoding but is generally more stringent.
The figure below shows a regime (panel D) where the Procrustes distance (x-axis) can be quite large even when the distance in terms of decoding is small (y-axis).
The figure below shows a regime (panel D) where the Procrustes distance (x-axis) can be quite large even when the distance in terms of decoding is small (y-axis).
Our contribution is to make this intuition mathematically precise.
Will spare you the details here, but the punchline is that CKA and other measures can be interpreted as measuring the *average agreement between decoders fit to two different systems*.
Teaser of the math below.
Will spare you the details here, but the punchline is that CKA and other measures can be interpreted as measuring the *average agreement between decoders fit to two different systems*.
Teaser of the math below.
November 18, 2024 at 6:18 PM
Our contribution is to make this intuition mathematically precise.
Will spare you the details here, but the punchline is that CKA and other measures can be interpreted as measuring the *average agreement between decoders fit to two different systems*.
Teaser of the math below.
Will spare you the details here, but the punchline is that CKA and other measures can be interpreted as measuring the *average agreement between decoders fit to two different systems*.
Teaser of the math below.
The basic punchline is very intuitive.
Neural representations with similar geometry will tend to produce similar predictions when given the same decoding task.
To the extent that geometry is (somewhat) different, decoded signals will also be (somewhat) different. Example below 👇
Neural representations with similar geometry will tend to produce similar predictions when given the same decoding task.
To the extent that geometry is (somewhat) different, decoded signals will also be (somewhat) different. Example below 👇
November 18, 2024 at 6:18 PM
The basic punchline is very intuitive.
Neural representations with similar geometry will tend to produce similar predictions when given the same decoding task.
To the extent that geometry is (somewhat) different, decoded signals will also be (somewhat) different. Example below 👇
Neural representations with similar geometry will tend to produce similar predictions when given the same decoding task.
To the extent that geometry is (somewhat) different, decoded signals will also be (somewhat) different. Example below 👇
My lab has been working on comparing neural representations for the past few years - methods like RSA, CKA, CCA, Procrustes distance
We are often asked: What do these things tell us about the system's function? How do they relate to decoding?
Our new paper has some answers arxiv.org/abs/2411.08197
We are often asked: What do these things tell us about the system's function? How do they relate to decoding?
Our new paper has some answers arxiv.org/abs/2411.08197
November 18, 2024 at 6:18 PM
My lab has been working on comparing neural representations for the past few years - methods like RSA, CKA, CCA, Procrustes distance
We are often asked: What do these things tell us about the system's function? How do they relate to decoding?
Our new paper has some answers arxiv.org/abs/2411.08197
We are often asked: What do these things tell us about the system's function? How do they relate to decoding?
Our new paper has some answers arxiv.org/abs/2411.08197