Robocrunch        AI

François Chollet   @fchollet

Deep learning @google. Creator of Keras. Author of 'Deep Learning with Python'. Opinions are my own.
fchollet 58.3

  Tweets by François Chollet  

François Chollet    @fchollet   ·   3 hours
Have you read the preview version of Deep Learning with Python, 2nd edition? Would you like to provide a quote we'll display on the back cover? Email me your quotes! (my email is on my website). Provide your name, your title & company/affiliation. Quotes should be 1-2 sentences.
 Reply      Retweet   5      Like     32    

François Chollet    @fchollet   ·   5 hours
New tutorial on training a deep learning model (MIRNet) to enhance low-light images -- with much nicer results than what you'd get from a simple auto-contrast correction.
 Reply      Retweet   30      Like     163    

François Chollet    @fchollet   ·   9/16/2021
New tutorial on, for audio data this time: implementing MelGAN, a vocoder architecture that can be used for speech synthesis.
 Reply      Retweet   2      Like     7    

François Chollet    @fchollet   ·   9/14/2021
Artificial restrictions on housing development are a regressive tax on the poor and a gift to the rich, period. It's wealth transfer from renters to owners.
 Reply      Retweet   2      Like     11    

François Chollet    @fchollet   ·   9/14/2021
The price of housing only goes up because of artificial constraints that prevent supply from meeting demand. Housing works as an investment because we don't build fast enough / dense enough.
 Reply      Retweet   3      Like     13    

François Chollet    @fchollet   ·   9/14/2021
Introducing TensorFlow Similarity: a toolbox to make it easy and fast to train similarity models.
 Reply      Retweet   3      Like     33    

François Chollet    @fchollet   ·   9/14/2021
New tutorial on Graph Attention Networks.
 Reply      Retweet   17      Like     83    

François Chollet    @fchollet   ·   9/14/2021
Captures a common pattern: many far-right/RETVRN folks like to exalt the Western cultural heritage, but in reality they don't appreciate it or even know it. It's just an ideological fashion statement for them. If they were into culture they'd push for funding education & the arts
 Reply      Retweet   5      Like     57    

François Chollet    @fchollet   ·   9/13/2021
The fact that the cost of housing in cities is growing faster than middle/working class income growth (due to artificial supply constraints) is one of the greatest threats to the prosperity of the middle/working class and one of the biggest drivers of inequality today.
 Reply      Retweet   3      Like     47    

François Chollet    @fchollet   ·   9/13/2021
New code example on the Swin Transformer architecture for image classification -- a Transformer-based vision model that uses local self-attention as a way to make self-attention on images linear in complexity.
 Reply      Retweet   11      Like     52    

François Chollet    @fchollet   ·   9/11/2021
How to spot an event that only occurs 0.18% of the time? Here's an essential technique that is used more rarely than it should: class weighting. A must for dealing with very imbalanced classification datasets, such as those you see in fraud detection.
 Reply      Retweet   18      Like     124    

François Chollet    @fchollet   ·   9/11/2021
The best parts of the internet are about individual empowerment: discover, learn, connect, do, build. The worst parts are about control of the individual: time-wasting, algorithmic information diets in which consumer has no say and that controls their beliefs.
 Reply      Retweet   21      Like     112    

François Chollet    @fchollet   ·   9/11/2021
I believe the main target of a good education should be a love of learning, confidence in one's ability to learn anything, and knowledge of how to learn effectively -- how to approach new domains and new problems. Education is the transition to autonomy. Not a molding process.
 Reply      Retweet   108      Like     538    

François Chollet    @fchollet   ·   9/9/2021
Idle observation -- TikTok virality dynamics open the door to entirely new categories of psychological magic tricks.
 Reply      Retweet        Like     16    

François Chollet    @fchollet   ·   9/6/2021
"To learn from data, you have to make assumptions about it. These assumptions define what can be learned. As such, the structure of your hypothesis space—the architecture of your model—is extremely important. It encodes [...] the prior knowledge that the model starts with" (DLwP)
 Reply      Retweet   1      Like     5    

François Chollet    @fchollet   ·   9/6/2021
Perhaps a very succinct way to put it is this: if it's a sound investment, knowledgeable actors are going to want to buy as much of it as possible. If it's a scam, knowledgeable actors are going to want *other people* to buy as much of it as possible.
 Reply      Retweet   3      Like     3    

François Chollet    @fchollet   ·   9/4/2021
A TensorFlow/Keras implementation of the additive attention Transformer paper (Wu et al) that was released a few days ago
 Reply      Retweet   46      Like     247    

François Chollet    @fchollet   ·   9/3/2021
In our example, we kept the data augmentation stage in float32 (created before setting the global policy) since it's meant to be run on CPU as part of the TF data pipeline.
 Reply      Retweet        Like     14    

François Chollet    @fchollet   ·   9/3/2021
Tweetorial: using mixed precision training with Keras models. All you need to do is set the dtype policy via `keras.mixed_precision.set_global_policy()`. Example: training an ImageNet model on Colab's TPUv2 (for free!) with bfloat16 mixed precision:
 Reply      Retweet   12      Like     83    

François Chollet    @fchollet   ·   9/2/2021
It's often the case that you can't understand a thing well purely by accumulating knowledge about it -- you need the context that's provided by knowing *other* things as well.
 Reply      Retweet   34      Like     235    

François Chollet    @fchollet   ·   8/31/2021
Also yes, this is exactly what the data says. It's an information ecosystem issue.
 Reply      Retweet   18      Like     70    

François Chollet    @fchollet   ·   8/29/2021
No need to invoke far-fetched psychological, cultural, or political theories. There are info ecosystems where vaccines are constantly decried as evil, while the miracle cure of the day is getting the spotlight. People who read this stuff every day believe it. How could they not?
 Reply      Retweet   6      Like     106    

François Chollet    @fchollet   ·   8/26/2021
Imagine a swarm of trillions of nanorobots that could assemble into a self-regenerating intelligent robot capable making copies of itself, exponentially... Yeah that's basically an ordinary multicellular organism
 Reply      Retweet   48      Like     465    

François Chollet    @fchollet   ·   8/24/2021
The greatest asset of a language or framework is its ecosystem, and bundling breaking changes in a single release invalidates large swathes of your ecosystem. Always go with incremental change, so you bring the ecosystem with you. Evolution, not mass extinction.
 Reply      Retweet   7      Like     41    

François Chollet    @fchollet   ·   8/21/2021
The dumbest, most impotent idea of the 2010s and 2020s is the notion that you can fight disinformation with actual information. It's like fighting a forest fire by planting trees.
 Reply      Retweet   74      Like     345    

François Chollet    @fchollet   ·   8/17/2021
Let me tell you about my favorite upcoming feature of TensorFlow/Keras 2.7: a better debugging experience. We've worked on 3 things that significantly cut the time it takes to analyze and fix issues in your Keras code.
 Reply      Retweet   123      Like     1027    

François Chollet    @fchollet   ·   7/26/2021
Two useful Keras patterns demonstrated in this example: 1. Create custom layers by subclassing the Layer class. 2. Writing custom training loops by overriding train_step() in the Model class. Let's take a look.
 Reply      Retweet   21      Like     150    

François Chollet    @fchollet   ·   7/20/2021
Most people I meet overestimate what deep learning can do (it's curve-fitting, don't expect it to do discrete symbol manipulation, it will solve symbolic tasks via embedding + interpolation) and simultaneously underestimate what you can do with curve-fitting given enough data
 Reply      Retweet   172      Like     1109    

François Chollet    @fchollet   ·   7/19/2021
Certain social networks willingly let themselves be used by bad actors as attack vectors on the minds of vulnerable people, with disastrous consequences (e.g. the anti-vax movement). Because it's profitable and they don't care. Not just in the US, but around the world
 Reply      Retweet   12      Like     116    

François Chollet    @fchollet   ·   7/19/2021
The human mind is highly vulnerable to certain patterns of social & psychological manipulation. Algorithm-powered social networks exploit these very patterns to maximize "engagement". In turn, bad actors exploit the algorithms to run sophisticated disinformation campaigns.
 Reply      Retweet   29      Like     183    

François Chollet    @fchollet   ·   7/19/2021
"Human nature has good & bad sides, and social media is just a reflection of human nature, so it includes the bad stuff" is not the wise take you think it is. Disinfo on social media is not an organic side-effect of free speech, it's the product of highly coordinated campaigns.
 Reply      Retweet   87      Like     323    

François Chollet    @fchollet   ·   7/16/2021
New code walkthrough on conditional GANs, for generating new images while controlling their appearance (e.g. by conditioning the generation process on a class).
 Reply      Retweet   40      Like     176    

François Chollet    @fchollet   ·   7/12/2021
Many runtimes other than Python TensorFlow understand the Keras graph-of-layers format, such as TF.js, CoreML, DeepLearning4J... A high level, human-readable saving format is much easier to implement for third-party platforms.
 Reply      Retweet   2      Like     21    

  Relevant People  

Yann LeCun
Professor at NYU. Chief AI Scientist at Facebook. Researcher in AI, Machine Learning, etc. ACM Turing Award Laureate.
Yann LeCun 58.4

François Chollet
Deep learning @google. Creator of Keras. Author of 'Deep Learning with Python'. Opinions are my own.
François Chollet 58.3

TensorFlow is a fast, flexible, and scalable open-source machine learning library for research and production.
TensorFlow 56.5

Facebook AI
Facebook AI focuses on bringing the world together by advancing AI, powering meaningful and safe experiences, and conducting open research.
Facebook AI 53.8

Hugging Face
The AI community building the future. #BlackLivesMatter #stopasianhate
Hugging Face 47.4

Sergey Levine
Associate Professor at UC Berkeley
Sergey Levine 43.3

Dmytro Mishkin
Marrying classical CV and Deep Learning. I do things, which work, rather than being novel, but not working.
Dmytro Mishkin 30.7