Emmanuel Soubies
@esoubies.bsky.social
CNRS researcher at IRIT.
Computational Imaging, Microscopy, Optimization, Machine Learning
Website: https://www.irit.fr/~Emmanuel.Soubies/
Computational Imaging, Microscopy, Optimization, Machine Learning
Website: https://www.irit.fr/~Emmanuel.Soubies/
Reposted by Emmanuel Soubies
Then why does flow matching generalize?? Because it fails!
The inductive bias of the neural network prevents from perfectly learning u* and overfitting.
In particular neural networks fail to learn the velocity field for two particular time values.
See the paper for a finer analysis 😀
The inductive bias of the neural network prevents from perfectly learning u* and overfitting.
In particular neural networks fail to learn the velocity field for two particular time values.
See the paper for a finer analysis 😀
June 18, 2025 at 8:15 AM
Then why does flow matching generalize?? Because it fails!
The inductive bias of the neural network prevents from perfectly learning u* and overfitting.
In particular neural networks fail to learn the velocity field for two particular time values.
See the paper for a finer analysis 😀
The inductive bias of the neural network prevents from perfectly learning u* and overfitting.
In particular neural networks fail to learn the velocity field for two particular time values.
See the paper for a finer analysis 😀
Seems that here a part of the signal of interest has been attenuated? From my own tests (with the Matlab version on SIM data), maybe this can be improved by playing with the threshold parameter.
May 16, 2025 at 2:20 PM
Seems that here a part of the signal of interest has been attenuated? From my own tests (with the Matlab version on SIM data), maybe this can be improved by playing with the threshold parameter.
In particular, it handles out-of-focus regions more effectively than traditional OTF attenuation.
May 15, 2025 at 12:31 PM
In particular, it handles out-of-focus regions more effectively than traditional OTF attenuation.