François Chollet Profile Banner
François Chollet Profile Image
François Chollet

@fchollet

Followers
231,053
Following
652
Media
831
Statuses
15,584

Deep learning @google. Creator of Keras. Author of 'Deep Learning with Python'. Opinions are my own.

United States
Joined August 2009
Don't wanna be here? Send us removal request.
Pinned Tweet
@fchollet
François Chollet
a year ago
I've just released a fairly lengthy paper on defining & measuring intelligence, as well as a new AI evaluation dataset, the "Abstraction and Reasoning Corpus". I've been working on this for the past 2 years, on & off. Paper: https://t.co/djNAIUZF7E ARC: https://t.co/MvubT2HTKT https://t.co/bVrmgLAYEv
Tweet media one
103
1K
5K
@fchollet
François Chollet
4 hours ago
It's not called Keras because making it is kerja keras... but it might as well be tbh ⚒️
11
18
78
@fchollet
François Chollet
20 hours ago
Most of the time, when we say something, what the other person ends up understanding is pretty different from what we intended to convey (including on Twitter, obviously). But it's still pretty amazing that there's any alignment at all. Language is a miracle.
16
47
409
@fchollet
François Chollet
23 hours ago
Take pride in what you do, not in an identity you were born into
13
109
786
@fchollet
François Chollet
a day ago
https://t.co/QFl5mdzgfN has now *64* code walkthroughs demonstrating common deep learning workflows -- a nice round number. I'm super proud of our contributor community that made it happen. Awesome contributions all around!
6
49
282
@fchollet
François Chollet
a day ago
The truth is in the code.
23
72
437
@fchollet
François Chollet
a day ago
Finally, it combines the encoder and decoder in a Model subclass, where training logic is packaged in the `train_step()` method (this enables training via `fit()`, which gives you callbacks and distribution support for free). Also note the `generate()` method for inference! https://t.co/cG6KkgUNNO
Tweet media one
Tweet media two
0
1
16
@fchollet
François Chollet
a day ago
Then it defines a Transformer encoder, which is your usual Transformer block, as well as a Transformer decoder, which is also your usual Transformer block, but with causal attention to prevent later timesteps to influence the decoding of earlier timesteps. https://t.co/Ige93alEwK
Tweet media one
Tweet media two
1
3
30
@fchollet
François Chollet
a day ago
This example was implemented by @NandanApoorv. Let's take a look at the model architecture. It starts by defining two embedding layers: a positional embedding for text tokens, and an embedding for speech features, that uses 1D convolutions with strides for downsampling. https://t.co/7bj9QtZSyV
Tweet media one
1
2
31
@fchollet
François Chollet
a day ago
New code walkthrough on https://t.co/m6mT8SrKDD: speech recognition with Transformer. Very readable and concise demonstration of how to build and train a speech recognition model on the LJSpeech dataset. https://t.co/LDKhOnIBLG
3
67
282
François Chollet Retweeted
@issielapowsky
issie lapowsky
a day ago
NEW: Between Aug 2020 and Jan 2021 far-right misinformation Facebook pages drew more engagement per follower than any other type of partisan news, a new study shows. These misinfo pages drew 65% more engagement per follower than even other far-right news. https://t.co/sk1vVNfH97
25
322
551
@fchollet
François Chollet
2 days ago
The speed at which I accumulate open tabs in Chrome is alarming. I have to declare tab bankruptcy every couple days
62
90
1K
@fchollet
François Chollet
3 days ago
Got this very nice book in the mail. All TensorFlow/Keras, with very readable code examples. Includes a section on StyleGAN, which will come in handy since I was trying to implement it the other day https://t.co/M56DB7E0Dm
Tweet media one
13
100
986
@fchollet
François Chollet
3 days ago
You can learn more about this pattern here: https://t.co/yzFys0aqlr
1
3
21
@fchollet
François Chollet
3 days ago
This example features a model with a custom `train_step()`. Overriding `train_step()` enables you to write arbitrary training logic (unsupervised clustering in this case) while benefiting from the features of `fit()`: callbacks, built-in distribution, step fusing on TPU... https://t.co/XUaIt0TFyo
Tweet media one
Tweet media two
2
5
32
@fchollet
François Chollet
3 days ago
New code walkthrough on https://t.co/m6mT8SrKDD: unsupervised image clustering with a contrastive loss. https://t.co/3huVnQ9pBP
2
32
173
François Chollet Retweeted
@PyImageSearch
Adrian Rosebrock
3 days ago
New tutorial!🚀 Adversarial attacks with FGSM (Fast Gradient Sign Method) - Implement FGSM with #Keras and #TensorFlow - Fool CNNs - Full tutorial w/ #Python code https://t.co/yAHUG9GBFf 👍 #DeepLearning #MachineLearning #ArtificialIngelligence #AI #DataScience #OpenCV https://t.co/Xb5O8ccbv8
Tweet media one
2
73
249