Jan van Gemert
@jvgemert.bsky.social
Head of the Computer Vision lab; TU Delft.
- Fundamental empirical Deep Learning research
- Visual inductive priors for data efficiency
Web: https://jvgemert.github.io/
- Fundamental empirical Deep Learning research
- Visual inductive priors for data efficiency
Web: https://jvgemert.github.io/
Isn't "training free" the current term for this?
November 22, 2025 at 5:35 PM
Isn't "training free" the current term for this?
It seems nice.. Why resist? 🤔
(I still make all my slides in Beamer 😅; I once tried remarkjs, which would allow animation, but I didn't get sufficiently proficient in it, and stayed in the latex env 😅)
(I still make all my slides in Beamer 😅; I once tried remarkjs, which would allow animation, but I didn't get sufficiently proficient in it, and stayed in the latex env 😅)
November 19, 2025 at 10:27 PM
It seems nice.. Why resist? 🤔
(I still make all my slides in Beamer 😅; I once tried remarkjs, which would allow animation, but I didn't get sufficiently proficient in it, and stayed in the latex env 😅)
(I still make all my slides in Beamer 😅; I once tried remarkjs, which would allow animation, but I didn't get sufficiently proficient in it, and stayed in the latex env 😅)
I really appreciate constructive initiatives and not "just complain" 🙂
I'll DM you for doing something constructive that I've been thinking about! 🙂
I'll DM you for doing something constructive that I've been thinking about! 🙂
November 17, 2025 at 9:28 AM
I really appreciate constructive initiatives and not "just complain" 🙂
I'll DM you for doing something constructive that I've been thinking about! 🙂
I'll DM you for doing something constructive that I've been thinking about! 🙂
Haha, very nicely done 🙂
November 17, 2025 at 8:17 AM
Haha, very nicely done 🙂
But it isn't a zero sum game (?)
November 17, 2025 at 8:15 AM
But it isn't a zero sum game (?)
I completely feel your frustration..
But.. I have been playing devil's advocate and saying:
Unless the papers are fraudulent, these papers are still producing knowledge.. despite such misaligned incentives, no?
But.. I have been playing devil's advocate and saying:
Unless the papers are fraudulent, these papers are still producing knowledge.. despite such misaligned incentives, no?
November 16, 2025 at 10:05 PM
I completely feel your frustration..
But.. I have been playing devil's advocate and saying:
Unless the papers are fraudulent, these papers are still producing knowledge.. despite such misaligned incentives, no?
But.. I have been playing devil's advocate and saying:
Unless the papers are fraudulent, these papers are still producing knowledge.. despite such misaligned incentives, no?
Why? :).
What operational, ie, significant, downstream difference will this difference make?
Making differences that make no difference is perhaps better left to the philosophers? They have no need of empiricism.
What operational, ie, significant, downstream difference will this difference make?
Making differences that make no difference is perhaps better left to the philosophers? They have no need of empiricism.
November 16, 2025 at 7:59 PM
Why? :).
What operational, ie, significant, downstream difference will this difference make?
Making differences that make no difference is perhaps better left to the philosophers? They have no need of empiricism.
What operational, ie, significant, downstream difference will this difference make?
Making differences that make no difference is perhaps better left to the philosophers? They have no need of empiricism.
What's the operational difference to call In-Context Learning a form of "learning" or not?
"A difference that makes no difference is no difference."
Platonification will not help, no?
Nearest Neighbors isn't "learning", but, so what? :)
"A difference that makes no difference is no difference."
Platonification will not help, no?
Nearest Neighbors isn't "learning", but, so what? :)
November 16, 2025 at 1:08 PM
What's the operational difference to call In-Context Learning a form of "learning" or not?
"A difference that makes no difference is no difference."
Platonification will not help, no?
Nearest Neighbors isn't "learning", but, so what? :)
"A difference that makes no difference is no difference."
Platonification will not help, no?
Nearest Neighbors isn't "learning", but, so what? :)
Oh wow! That looks wonderful!
I wanted to do the same, but the rain, well, rained on my parade 😅
I wanted to do the same, but the rain, well, rained on my parade 😅
November 14, 2025 at 4:48 PM
Oh wow! That looks wonderful!
I wanted to do the same, but the rain, well, rained on my parade 😅
I wanted to do the same, but the rain, well, rained on my parade 😅
Interesting question! Computer vision system diagrams seem mainly colored boxes with arrows (?)
November 10, 2025 at 6:37 AM
Interesting question! Computer vision system diagrams seem mainly colored boxes with arrows (?)
The new generation of computer vision researchers might think that way about the GPUs used by Alex Krizhevsky 😉.
Are the original GPUs still around somewhere? 🤔
Are the original GPUs still around somewhere? 🤔
October 28, 2025 at 6:17 PM
The new generation of computer vision researchers might think that way about the GPUs used by Alex Krizhevsky 😉.
Are the original GPUs still around somewhere? 🤔
Are the original GPUs still around somewhere? 🤔
I'd be very interested in the longer piece.
Because.. In my opinion, at the moment this thread is too much about publicly shaming people, and not enough about arguing against the claims made.
Because.. In my opinion, at the moment this thread is too much about publicly shaming people, and not enough about arguing against the claims made.
October 27, 2025 at 3:52 PM
I'd be very interested in the longer piece.
Because.. In my opinion, at the moment this thread is too much about publicly shaming people, and not enough about arguing against the claims made.
Because.. In my opinion, at the moment this thread is too much about publicly shaming people, and not enough about arguing against the claims made.
Amazing work! This reminds me of adversarial attacks (optimize the confidence) but also of DeepDream (maximize confidence).
Also the Deep Image Prior comes to mind; ie: the bias in the architecture 🙂
If I may ask, for the brain, do you think the priors are learned? Or hard-coded (evolution)? 🤔
Also the Deep Image Prior comes to mind; ie: the bias in the architecture 🙂
If I may ask, for the brain, do you think the priors are learned? Or hard-coded (evolution)? 🤔
October 26, 2025 at 4:22 PM
Amazing work! This reminds me of adversarial attacks (optimize the confidence) but also of DeepDream (maximize confidence).
Also the Deep Image Prior comes to mind; ie: the bias in the architecture 🙂
If I may ask, for the brain, do you think the priors are learned? Or hard-coded (evolution)? 🤔
Also the Deep Image Prior comes to mind; ie: the bias in the architecture 🙂
If I may ask, for the brain, do you think the priors are learned? Or hard-coded (evolution)? 🤔
We created metascience for ML, metascienceforml.github.io also for PhD candidates, and a MSc course 🙂
Metascience for Machine Learning
A repository for collecting information about Metascience for machine learning.
metascienceforml.github.io
October 26, 2025 at 7:25 AM
We created metascience for ML, metascienceforml.github.io also for PhD candidates, and a MSc course 🙂
Ben Recht @beenwrekt.bsky.social has a brilliant blog discussing, among other things, how significant those stats are 🙂.
www.argmin.net
www.argmin.net
arg min | Ben Recht | Substack
arg min: a blog of minimum value. on the history, foundations, and validity of "optimally" automated decision making. Click to read arg min, by Ben Recht, a Substack publication with thousands of subs...
www.argmin.net
October 23, 2025 at 8:54 AM
Ben Recht @beenwrekt.bsky.social has a brilliant blog discussing, among other things, how significant those stats are 🙂.
www.argmin.net
www.argmin.net
Classicism, as measured in years: max Before Alexnet (or min After Marr) 😉
October 22, 2025 at 10:12 AM
Classicism, as measured in years: max Before Alexnet (or min After Marr) 😉
How so? 🤔.
And how do we do Classical time keeping (?)
- 10 years BA (Before Alexnet)
- 10 years AM (After Marr)
- ... (?)
And how do we do Classical time keeping (?)
- 10 years BA (Before Alexnet)
- 10 years AM (After Marr)
- ... (?)
October 21, 2025 at 8:16 PM
How so? 🤔.
And how do we do Classical time keeping (?)
- 10 years BA (Before Alexnet)
- 10 years AM (After Marr)
- ... (?)
And how do we do Classical time keeping (?)
- 10 years BA (Before Alexnet)
- 10 years AM (After Marr)
- ... (?)
MNIST suddenly became the real world 😉
Yann knew this all along!
Yann knew this all along!
October 21, 2025 at 6:20 PM
MNIST suddenly became the real world 😉
Yann knew this all along!
Yann knew this all along!
- Perdido street station
- Heinlein's books, if you haven't already 🙂
- Daemon
- Dragon's egg (yes, scifi, not fantasy 🙂)
Peter Watts I also recommend 🙂
- Heinlein's books, if you haven't already 🙂
- Daemon
- Dragon's egg (yes, scifi, not fantasy 🙂)
Peter Watts I also recommend 🙂
October 20, 2025 at 8:06 PM
- Perdido street station
- Heinlein's books, if you haven't already 🙂
- Daemon
- Dragon's egg (yes, scifi, not fantasy 🙂)
Peter Watts I also recommend 🙂
- Heinlein's books, if you haven't already 🙂
- Daemon
- Dragon's egg (yes, scifi, not fantasy 🙂)
Peter Watts I also recommend 🙂
- Speed of dark (related to Flowers for Algernon)
- long way to a small angry planet
- Klara and the Sun
- A psalm for the wild built, loved this robot tale
- Anathem, starts slow, but ends 🤯
- A wild sheep chase, by Murakami
- The science of Discworld, I liked a lot, albeit "science fantasy" 😊
- long way to a small angry planet
- Klara and the Sun
- A psalm for the wild built, loved this robot tale
- Anathem, starts slow, but ends 🤯
- A wild sheep chase, by Murakami
- The science of Discworld, I liked a lot, albeit "science fantasy" 😊
October 20, 2025 at 7:56 PM
- Speed of dark (related to Flowers for Algernon)
- long way to a small angry planet
- Klara and the Sun
- A psalm for the wild built, loved this robot tale
- Anathem, starts slow, but ends 🤯
- A wild sheep chase, by Murakami
- The science of Discworld, I liked a lot, albeit "science fantasy" 😊
- long way to a small angry planet
- Klara and the Sun
- A psalm for the wild built, loved this robot tale
- Anathem, starts slow, but ends 🤯
- A wild sheep chase, by Murakami
- The science of Discworld, I liked a lot, albeit "science fantasy" 😊
😂 I'll need to come over during cvpr, and practice that! 😊
October 19, 2025 at 12:20 PM
😂 I'll need to come over during cvpr, and practice that! 😊
I always pronounce Lenet with a French accent in my Deep Learning MSc course (400 students). And I tell them it's mandatory for the exam...😊
October 18, 2025 at 6:52 PM
I always pronounce Lenet with a French accent in my Deep Learning MSc course (400 students). And I tell them it's mandatory for the exam...😊
How do I pronounce CVPR with a French accent? 🤔
October 18, 2025 at 6:50 PM
How do I pronounce CVPR with a French accent? 🤔