Peyman Milanfar
@docmilanfar.bsky.social
Distinguished Scientist at Google. Computational Imaging, Machine Learning, and Vision. Posts are personal opinions. May change or disappear over time.
http://milanfar.org
http://milanfar.org
I was invited to give this plenary talk at the Electronic Imaging Symposium this past February. The recording is now available on YouTube
www.youtube.com/watch?v=s59U...
www.youtube.com/watch?v=s59U...
April 14, 2025 at 12:57 AM
I was invited to give this plenary talk at the Electronic Imaging Symposium this past February. The recording is now available on YouTube
www.youtube.com/watch?v=s59U...
www.youtube.com/watch?v=s59U...
Silicon Valley crosswalk buttons have been hacked to hilarious effect
April 13, 2025 at 4:32 AM
Silicon Valley crosswalk buttons have been hacked to hilarious effect
ruined by tariffs
April 12, 2025 at 9:15 PM
ruined by tariffs
bias-variance tradeoff
April 11, 2025 at 12:00 PM
bias-variance tradeoff
Autoregressive models have become more popular recently. Speaking to people, I get the sense most folks are unaware of the history of the topic in signal processing. In particular, AR across scale got a lot of attention in the 90s. There are a lot of great papers on the subject but one stand out…..
April 3, 2025 at 3:52 AM
Autoregressive models have become more popular recently. Speaking to people, I get the sense most folks are unaware of the history of the topic in signal processing. In particular, AR across scale got a lot of attention in the 90s. There are a lot of great papers on the subject but one stand out…..
give this kid a PhD
March 28, 2025 at 3:30 AM
give this kid a PhD
Model Distillation
March 24, 2025 at 3:12 AM
Model Distillation
And.... you have Tweedie's formula for the MMSE denoiser in explicit form. Specifically, it's the sum of a "trivial" max likelihood estimate; and a Bayes "correction"
6/n
6/n
March 18, 2025 at 6:12 AM
And.... you have Tweedie's formula for the MMSE denoiser in explicit form. Specifically, it's the sum of a "trivial" max likelihood estimate; and a Bayes "correction"
6/n
6/n
Now divide both sides by P(y). This is not problematic since P(y) - being a Gaussian "blurred" version of the prior P(x) - does not vanish.
5/n
5/n
March 18, 2025 at 6:12 AM
Now divide both sides by P(y). This is not problematic since P(y) - being a Gaussian "blurred" version of the prior P(x) - does not vanish.
5/n
5/n
The key to arriving at Tweedie's formula is to differentiate the marginal density P(y).
4/n
4/n
March 18, 2025 at 6:12 AM
The key to arriving at Tweedie's formula is to differentiate the marginal density P(y).
4/n
4/n
Note that the normalizing term in the denominator is the marginal density of y.
3/n
3/n
March 18, 2025 at 6:12 AM
Note that the normalizing term in the denominator is the marginal density of y.
3/n
3/n
The MMSE denoiser is known to be the conditional mean f̂(y) = 𝔼(x|y). In this case, we can write the expression for this conditional mean explicitly:
2/n
2/n
March 18, 2025 at 6:12 AM
The MMSE denoiser is known to be the conditional mean f̂(y) = 𝔼(x|y). In this case, we can write the expression for this conditional mean explicitly:
2/n
2/n
Tweedie's formula is super important in diffusion models & is also one of the cornerstones of empirical Bayes methods.
Given how easy it is to derive, it's surprising how recently it was discovered ('50s). It was published a while later when Tweedie wrote Stein about it
1/n
Given how easy it is to derive, it's surprising how recently it was discovered ('50s). It was published a while later when Tweedie wrote Stein about it
1/n
March 18, 2025 at 6:12 AM
Tweedie's formula is super important in diffusion models & is also one of the cornerstones of empirical Bayes methods.
Given how easy it is to derive, it's surprising how recently it was discovered ('50s). It was published a while later when Tweedie wrote Stein about it
1/n
Given how easy it is to derive, it's surprising how recently it was discovered ('50s). It was published a while later when Tweedie wrote Stein about it
1/n
1st one that we know of
March 15, 2025 at 9:21 PM
1st one that we know of
really, how can you
March 15, 2025 at 3:03 AM
really, how can you
The tech industry in Silicon Valley runs on the considerable talents and work ethic of folks who were not born here. Many of these folks were also trained in this country in our Universities. This is our strength.
If we mess with this virtuous cycle, it is at our own peril.
If we mess with this virtuous cycle, it is at our own peril.
March 12, 2025 at 4:00 AM
The tech industry in Silicon Valley runs on the considerable talents and work ethic of folks who were not born here. Many of these folks were also trained in this country in our Universities. This is our strength.
If we mess with this virtuous cycle, it is at our own peril.
If we mess with this virtuous cycle, it is at our own peril.
It was published a couple of decades after the original (classified) work. Book lays out principles for how to effectively allocate search effort to maximize the probability of finding a target, with limited resources.
To my knowledge, it's not known or cited by RL folks
2/2
To my knowledge, it's not known or cited by RL folks
2/2
March 11, 2025 at 5:38 AM
It was published a couple of decades after the original (classified) work. Book lays out principles for how to effectively allocate search effort to maximize the probability of finding a target, with limited resources.
To my knowledge, it's not known or cited by RL folks
2/2
To my knowledge, it's not known or cited by RL folks
2/2
In the 50s a sophisticated theory of how to look for things was developed. Rooted in operations research, it was largely motivated by the loss of an H-bomb and a nuclear submarine in the middle of the ocean
Closely related to concepts in reinforcement learning
1/2
Closely related to concepts in reinforcement learning
1/2
March 11, 2025 at 5:38 AM
In the 50s a sophisticated theory of how to look for things was developed. Rooted in operations research, it was largely motivated by the loss of an H-bomb and a nuclear submarine in the middle of the ocean
Closely related to concepts in reinforcement learning
1/2
Closely related to concepts in reinforcement learning
1/2
Have you said thank you once to Tweedie?
March 4, 2025 at 5:58 AM
Have you said thank you once to Tweedie?
Einstein’s eulogy of the great Emmy Noether (1882-1935)
“In the judgment of the most competent living mathematicians, Fraulein Noether was the most significant creative mathematical genius thus far produced since the higher education of women began”
“In the judgment of the most competent living mathematicians, Fraulein Noether was the most significant creative mathematical genius thus far produced since the higher education of women began”
February 26, 2025 at 7:16 AM
Einstein’s eulogy of the great Emmy Noether (1882-1935)
“In the judgment of the most competent living mathematicians, Fraulein Noether was the most significant creative mathematical genius thus far produced since the higher education of women began”
“In the judgment of the most competent living mathematicians, Fraulein Noether was the most significant creative mathematical genius thus far produced since the higher education of women began”
meet Chris J Li - this titan of thought has single-handedly conquered the fields of machine learning, optimization, statistics, reinforcement learning, and federated learning.
he's not the visionary we want, but judging by the current state of affairs, he may be the one we deserve
he's not the visionary we want, but judging by the current state of affairs, he may be the one we deserve
February 23, 2025 at 7:19 AM
meet Chris J Li - this titan of thought has single-handedly conquered the fields of machine learning, optimization, statistics, reinforcement learning, and federated learning.
he's not the visionary we want, but judging by the current state of affairs, he may be the one we deserve
he's not the visionary we want, but judging by the current state of affairs, he may be the one we deserve
Reinforcement Learning
February 5, 2025 at 3:22 AM
Reinforcement Learning
excellent use of negative space
February 1, 2025 at 2:38 AM
excellent use of negative space
Prof. on website vs Prof. in person
January 15, 2025 at 4:22 AM
Prof. on website vs Prof. in person