Fun style transfer in the browser using MobileNet. No cloud component, runs entirely in the browser.
— Reza Zadeh (@Reza_Zadeh) November 20, 2018
Live Demo: https://t.co/0wcRH26ocW
Code: https://t.co/BZ2qDShMhN
Paper: https://t.co/wnmgKRHqow pic.twitter.com/asfui0JuNA
Our new #Corl18 paper:
— Pierre-Yves Oudeyer (@pyoudeyer) November 19, 2018
How curiosity-driven autonomous goal setting enables to discover
independantly controllable features of environments
pdf: https://t.co/vArIwNgewY
Blog: https://t.co/7LV1fs67NX
Colab: https://t.co/RIJ6GP9vTV#machinelearning #AI #NeuralNetworks
Two excellent papers building on each other to show:
— Reza Zadeh (@Reza_Zadeh) November 16, 2018
1) The loss of two-layer neural networks can be optimized to zero in polynomial time using gradient descent: https://t.co/pkvzNMH7hX
2) Same for ResNet, teasing out benefits of residual connections: https://t.co/ZfS1eW4us8 pic.twitter.com/5Tb5m0fia8
Have been experimenting with an unsupervised pre-training technique for @spacy_io, similar to ULMFit/Elmo/BERT etc. Still very early, but pushing to spacy-nightly soon so you can check it out! 🚀💫https://t.co/6lqeUj2uKj
— Matthew Honnibal (@honnibal) November 15, 2018
Neat idea: dropout where you randomly set some of the nonlinearities to being linear.
— Thomas Lahore (@evolvingstuff) November 15, 2018
Drop-Activation: Implicit Parameter Reduction and Harmonic Regularizationhttps://t.co/09rlgo1fom pic.twitter.com/kz4NLObcQv
Remember Lord's paradox (#Bookofwhy, chpt 6)?? Stephen Senn just posted a "statistical" solution here: https://t.co/MsWxaqv7OY to which I have added a commentary showing how it ought to be solved in our millennium. Enjoy the contrast. #causalinference
— Judea Pearl (@yudapearl) November 12, 2018
The BigGAN generators from our paper https://t.co/QUYlE9IBsE are now available on TF Hub (https://t.co/GHM9pIgQPw). Try the Colab demo at: https://t.co/Ynyb9T9AAD
— DeepMind (@DeepMindAI) November 12, 2018
Deep Learning exploded when open tools made it *easy*
— Trask (@iamtrask) November 12, 2018
I'm extremely excited to share our new @NipsConference paper feat:
- Federated Learning
- Differential Privacy
- Encrypted Data/Models#privacy made... *easy* 🥳
Paper: https://t.co/yLwpjiXXMX
Code: https://t.co/gKzzfw0mIV pic.twitter.com/iJ29MzHN86
So much wisdom in this retrospective on AirBnB's use of neural nets for search ranking: https://t.co/D8U3CbbMNc
— Pete Warden (@petewarden) November 10, 2018
"moving to deep learning is not a drop-in model replacement at all; rather it’s about scaling the system"
Entertaining read.
— Ferenc Huszár🇪🇺 (@fhuszar) November 9, 2018
I especially like the "failed models" and "feature engineering" sections.
NNs not requiring feature engineering is very much a myth in general. It is mainly true for certain dense high-D data.https://t.co/dZbq1LoaQk
Slowly but surely, deep learning is joining the last domain to resist it, tabular data. The networks are still relatively simple which makes me think we need better primitives for structured data. What is the convolution equivalent for spreadsheets? https://t.co/9gjCcNOdsG pic.twitter.com/4DBDrSBoeQ
— Emmanuel Ameisen (@EmmanuelAmeisen) November 9, 2018
WaveGlow: A Flow-based Generative Network for Speech Synthesis
— ML Review (@ml_review) November 9, 2018
By NVIDIA (@DeepPrenger @RafaelValleArt @ctnzr)
Generates HQ speech from mel-spectrograms 25x faster than real time on V100 GPU
PyTorch + Pretrainedhttps://t.co/IzCHtPjow0
ArXivhttps://t.co/HlVltjrZzl pic.twitter.com/VtIZWBE9sl
