Deep convolutional neural networks are great models of the visual system, but these static systems don't explain the temporal dynamics of real visual responses. So we built deep recurrent networks:@aran_nayebi @qbilius @SussilloDavid @NeuroAILab
— Daniel Bear (@recursus) July 3, 2018
Paper: https://t.co/QpheOUU7Dl pic.twitter.com/veyXeg5RGi
Dropout has been a standard technique for years, but we’re still finding new ways to interpret it.
— Denny Britz (@dennybritz) July 2, 2018
I tend to think of new techniques/architectures as an advancement in either regularization or optimization/gradient flow, but sometimes it’s difficult to tell which one it is. https://t.co/74lt0VtExG
While DropOut has been considered to prevent the co-adaption among hidden neurons to regularize the model, DropOut actually makes gradient flows even when the activation functions are saturated and helps the optimization converges to the flat minima. https://t.co/YEH33wWqRZ
— Daisuke Okanohara (@hillbig) July 2, 2018
Nice summary! It seems I missed to check some papers... https://t.co/e0jAn6gYFk
— Jin-Hwa Kim (@jnhwkim) July 1, 2018
Top 20 Deep Learning Papers, 2018 Edition https://t.co/xCjXFBc12v pic.twitter.com/6FAc4DLxcw
— KDnuggets (@kdnuggets) July 1, 2018
sounds relatively similar to permutation importance, but with removal instead of permutation (https://t.co/5fyL81kL1o)
— Sebastian Raschka (@rasbt) June 30, 2018
A toy example to illustrate the decision boundaries learned using a narrow vanilla fully connected network vs a narrow ResNet. pic.twitter.com/v8hmEYhIpJ
— hardmaru (@hardmaru) June 30, 2018
ResNet with one-neuron hidden layers is a Universal Approximator: The main contribution of this paper is to show that ResNet with one single neuron per hidden layer is enough to provide universal approximation as the depth goes to infinity. https://t.co/dujgHIVrbg pic.twitter.com/G9zOojqhtB
— hardmaru (@hardmaru) June 30, 2018
Image generation from scene graphs!! Check out our codes here! https://t.co/4BNQngOFyq
— Fei-Fei Li (@drfeifei) June 29, 2018
Important work from @sarahookr, @doomie, @piekindermans, and @_beenkim on quantifying the extent to which various saliency methods *actually* find relevant portions of images.
— Ari Morcos (@arimorcos) June 29, 2018
Really happy to see more work towards bringing rigor into interpretability! https://t.co/AHvi2N3Ods pic.twitter.com/kSMONsOFB2
👩🏫📃 This is such a fun paper!#DeepMoji: Using millions of emoji occurrences to learn any-domain representations for detecting sentiment, emotion and sarcasm. (@MIT @MediaLab, 2017). Built on a corpus of 1246 million tweets and 64 common emojis!https://t.co/YpGorr3Wr8 pic.twitter.com/PWa0z3rfbV
— 👩💻 DynamicWebPaige @ $HOME 🏞 (@DynamicWebPaige) June 29, 2018
The great thing about this paper is how they made the results and method completely open, and we can easily try it out using the code in the colab notebook on other tasks: https://t.co/5Vh9lyAwOK
— hardmaru (@hardmaru) June 28, 2018
