Tensors and matrices are the building blocks of ML models. Learn about the different techniques for compressing matrices in this post about Matrix Compression Operators!
— TensorFlow (@TensorFlow) February 15, 2020
🛠Read the blog → https://t.co/CvJv3PE2Ro pic.twitter.com/z6LEpAQxlD
Here's a lecture by Vladimir Vapnik on the Complete Statistical Theory of Learning, as part of the MIT Deep Learning lecture series. It is at times math-intensive and at times philosophical, culminating in a challenge to solve MNIST with very few examples. https://t.co/90gconCwMI pic.twitter.com/fQdySAtUbw
— Lex Fridman (@lexfridman) February 15, 2020
Time to get hacking! @huggingface just released an NPM package for question answering using #DistilBERT directly in #NodeJS with a 2x performance boost over Python. #madeWithTFJS #JavaScript https://t.co/oAB4j2ektr
— TensorFlow (@TensorFlow) February 14, 2020
The Deep Learning Compiler: A Comprehensive Survey [36pp]
— ML Review (@ml_review) February 14, 2020
Dissect general design of existing DL compilers & analyses intermediate representations, compiler optimization & much morehttps://t.co/552QG4rml7 pic.twitter.com/BvTCDEPXXH
2/5 Knowledge distillation by @geoffreyhinton @OriolVinyalsML @JeffDean originally motivated to transfer knowledge from large to smaller networks. Self-distillation is special case with identical architectures; predictions of model are fed back to itself as new target values.
— Hossein Mobahi (@TheGradient) February 14, 2020
1/5 Self-Distillation loop (feeding predictions as new target values & retraining) improves test accuracy. But why? We show it induces a regularization that progressively limits # of basis functions used to represent the solution. https://t.co/570qXFmlGj w/@farajtabar P.Bartlett pic.twitter.com/b79Q6ZSxlS
— Hossein Mobahi (@TheGradient) February 14, 2020
Neat tree of dplyr actions by @alexcourtiol & @rdataberlin:
— Mara Averick (@dataandme) February 10, 2020
📑 "Rguides: Data transformation with {dplyr}" https://t.co/g6ECT9uGov #rstats pic.twitter.com/uFtaSbaayQ
My slides: https://t.co/BZZgcTW4jO
— Yann LeCun (@ylecun) February 10, 2020
Video of the talks: https://t.co/Npou9MLZsA
- 1:10 in for Geoff Hinton's keynote,
- 1:44 for mine,
- 2:18 for Yoshua Bengio's,
- 2:51 for the panel discussion moderated by Leslie Pack Kaelbling
"Upgrading Python lists" by Dipam Vasani in @TDataScience
— Jeremy Howard (@jeremyphoward) February 7, 2020
First tutorial on fastcore's `L` class that I've seen. Check it out - it's a really helpful class and I use it ALL THE DAMN TIME nowadays!
NB: you don't need fastai2 for this, only fastcore,.https://t.co/9AH3HukzIv
I suppose we should actually write a damn paper about fastai.tabular sometime...
— Jeremy Howard (@jeremyphoward) February 7, 2020
Maybe we should try to trick @TheZachMueller into writing it. https://t.co/0mhX8uQjRo
And today I've tried UMAP with @nvidia @rapidsai in @kaggle kernels. A speedup of 120 x is nothing to sneeze at, even though it's not as dramatic as the 2000 x speedup for t-SNE.https://t.co/iIA8UZekH5 pic.twitter.com/QVRdqJQWag
— Bojan Tunguz (@tunguz) February 6, 2020
I just compared t-SNE algorithm on #MNIST dataset in @kaggle kernels between #sklearn and @rapidsai. We are getting a 2000X speedup!https://t.co/pbFgCbQ7jZ
— Bojan Tunguz (@tunguz) February 6, 2020
