
Now TensorFlow.js. a hardware accelerated machine intelligence library for the web. Official #deeplearnjs twitter, run by @nsthorat and @dsmilkov.
Cambridge, MA
Joined August 2017
- Tweets75
- Following33
- Followers3K
- Likes50
deeplearn.js v0.5 released! Major API simplifications, eager mode, chaining, and brand new interactive API reference.
Release notes: github.com/PAIR-code/deeplea…
API reference: deeplearnjs.org/docs/api/
Send us feedback!
Want to help out with deeplearn.js? Check out this issue which enumerates which ops we need gradients for in the new eager mode:
github.com/PAIR-code/deeplea…
Check out deeplearn-gl, a fork of headless-gl by @danfoody: headless deeplearn.js running with node.js & OpenGL!
github.com/dfoody/headless-g…
Don't forget to check out the interactive latent space explorer for the fonts model in the new distill post on augmenting human intelligence by @shancarter+@bengiswex.
We invite the OSS community to extend this UI to other generative deeplearn.js models.
deeplearnjs.org/demos/latent…
Performance RNN + Disklavier player piano + Input Conditioning + deeplearn.js duet
youtube.com/watch?v=-V_ulHS1…
~108ms on a mid-2015 MacbookPro
~19.88 GFLOPS
only ~3.4 times faster (and we're not accounting for download time in Chrome so this is an upper bound)
We still think there are huge performance wins from changing memory layout. Expect these numbers only to improve!
iPhone X // dljs benchmark:
~370ms to multiply two 1024D square matrices and download result, O(N^3).
~5.80 GFLOPS lower bound (can’t account for download and memory IO)
Cam Arcade: Turn your webcam into a controller and play old games like DOOM with deeplearn.js, by @bengiswex.
We are also releasing a JavaScript module for image classification - the first of our deeplearn.js components. Check out the code for more!
deeplearnjs.org/demos/teacha…
Tinker with deeplearn.js in browser with our new interactive playground!
Sharing coming soon with github gists.
deeplearnjs.org/demos/playgr…
A sneak preview of the model builder training MNIST on an iPhone 6 in Safari. deeplearnjs 0.3.0 with iOS support coming soon!
Version 0.2 released!
Broadcasting, LSTM inference, TF-like conv API, performance profiling in debug mode & more
github.com/PAIR-code/deeplea…
We just added Fashion MNIST to the model builder demo!
Build a new model without any code here:
pair-code.github.io/deeplear…










