My latest anomaly detection paper using LOO KDE with @SevvandiK, and an associated R package lookout: https://t.co/uTa99PuOmS pic.twitter.com/dASPV0Jn2h
— Rob J Hyndman (@robjhyndman) February 7, 2021
My latest anomaly detection paper using LOO KDE with @SevvandiK, and an associated R package lookout: https://t.co/uTa99PuOmS pic.twitter.com/dASPV0Jn2h
— Rob J Hyndman (@robjhyndman) February 7, 2021
Why does Stochastic Gradient Descent generalise well in deep networks?
— DeepMind (@DeepMind) February 3, 2021
Our team shows that if the learning rate is small but finite, the mean iterate of random shuffling SGD stays close to the path of gradient flow, but on a modified loss landscape https://t.co/JUAzPujWfP pic.twitter.com/NtlbVaMfLC
Scaling Laws for Transferhttps://t.co/m7Vh5aOTjx pic.twitter.com/5lw9GyOAp5
— hardmaru (@hardmaru) February 3, 2021
Turns out we can reverse engineer chunks of neural network and then write out weights by hand that reimplement it. Not sure what higher standard there is for showing we understand something. https://t.co/JKcpEolUJu
— Chris Olah (@ch402) February 2, 2021
Large, natural datasets are invaluable for training accurate, deployable systems, but are they required for driving modeling innovation? Can we use small, synthetic benchmarks instead? Our new paper asks this: https://t.co/b9WYYonQxi
— Nelson Liu (@nelsonfliu) February 2, 2021
w/ Tony Lee, @robinomial, @percyliang
(1/8) pic.twitter.com/sdhRk5UT5L
Speech Recognition by Simply Fine-tuning BERT
— AK (@ak92501) February 2, 2021
pdf: https://t.co/2kit83mnj9
abs: https://t.co/qyOosTp8Ey pic.twitter.com/brIQfVxWim
My awesome colleagues have now released #PyTorch version of StyleGAN2-ADA. (The initial release was in #TensorFlow )
— Ming-Yu Liu (@liu_mingyu) February 1, 2021
ADA uses a clever data augmentation to help address limit sample problems in #GAN training.https://t.co/75Yttri2KS
Introducing TT-Rec, a new method to dramatically shrink memory-intensive Deep Learning Recommendation Models and make them easier to deploy at scale. https://t.co/IiGmIakkZZ pic.twitter.com/pjtGhZhOG3
— Facebook AI (@facebookai) January 29, 2021
There has been much interest in ML methods that generate source code (e.g. Python) from English commands. But does this actually help software developers? We asked 31 developers to use a code generation plugin, and found some interesting results: https://t.co/ifiG3EYK3J 1/7 pic.twitter.com/u0mgY0LSSj
— Graham Neubig (@gneubig) January 28, 2021
"The results ... suggest that decision-makers can actually rid themselves of guilt more easily by delegating to machines than by delegating to other people." 😬
— Miles Brundage (@Miles_Brundage) January 28, 2021
"Hiding Behind Machines: When Blame Is Shifted to Artificial Agents," Feier et al.: https://t.co/fpNTNdkY7A
Want Transformers to perform long chains of reasoning and to remember stuff?
— Yann LeCun (@ylecun) January 27, 2021
Use Feedback Transformers.
Brought to you by a team from FAIR-Paris.@tesatory https://t.co/dpCzt2yiPf
what a simple yet effective idea! :)
— Kyunghyun Cho (@kchonyc) January 27, 2021
looking at it from the architectural depth perspective (https://t.co/DM4MvzqFQW by zheng et al.,) the depth (# of layers between a particular input at time t' and output at time t) is now (t-t') x L rather than (t-t') + L. https://t.co/GkkuMH7a5u pic.twitter.com/eXWeqoEKxx