ALADIN: All Layer Adaptive Instance Normalization for Fine-grained Style Similarity
— AK (@ak92501) March 18, 2021
pdf: https://t.co/K7homQW2GM
abs: https://t.co/jTp3DS4u7o pic.twitter.com/NUBfHbZZWE
ALADIN: All Layer Adaptive Instance Normalization for Fine-grained Style Similarity
— AK (@ak92501) March 18, 2021
pdf: https://t.co/K7homQW2GM
abs: https://t.co/jTp3DS4u7o pic.twitter.com/NUBfHbZZWE
Is it Enough to Optimize CNN Architectures on ImageNet?
— AK (@ak92501) March 17, 2021
pdf: https://t.co/zC5jToLTto
abs: https://t.co/oIYWstLrIf pic.twitter.com/RrEHLNJEqa
Multi-view subword regularization is simple but yields consistent improvements over pre-trained multilingual models. The best thing: It only needs to be applied during fine-tuning.
— Sebastian Ruder (@seb_ruder) March 16, 2021
Paper: https://t.co/gxTgbzVvWN
Code: https://t.co/FqUyZgEnOQ https://t.co/sTFxot6yan
Few-Shot Text Classification with Triplet Networks, Data Augmentation, and Curriculum Learning
— AK (@ak92501) March 16, 2021
pdf: https://t.co/uyzxL1tu4L
abs: https://t.co/qX90yC7nAH pic.twitter.com/Y15zxoaJo8
Revisiting ResNets: Improved Training and Scaling Strategies
— AK (@ak92501) March 16, 2021
pdf: https://t.co/Pn5cU2SVkB
abs: https://t.co/icpnuFwmXU pic.twitter.com/bA0E1GWR5z
Facebook AI has built TimeSformer, a new architecture for video understanding. It’s the first based exclusively on the self-attention mechanism used in Transformers. It outperforms the state of the art while being more efficient than 3D ConvNets for video.https://t.co/8mQ2rMgcDo pic.twitter.com/dBpbT3UJRx
— Facebook AI (@facebookai) March 15, 2021
A blog post on algorithmic fairness work at Facebook,
— Yann LeCun (@ylecun) March 11, 2021
and a research paper on the topic.
Paper: "Fairness On The Ground" https://t.co/InVLlshw7Q
Blog post: "What AI fairness in practice looks like at Facebook"https://t.co/sUmXKyAYu9
Pretrained Transformers as Universal Computation Engines https://t.co/dsmqvAEDCB
— /MachineLearning (@slashML) March 11, 2021
Check out our @googleai blog post on new framework to understand generalization in deep learning. paper: https://t.co/4piRRt2zHI with @bneyshabur, @PreetumNakkiran https://t.co/Wlbk2cH4nz
— Hanie Sedghi (@HanieSedghi) March 10, 2021
We introduce a new approach for image compression: instead of storing the pixels in an image, we store the weights of an MLP overfitted to the image 🌟 At low bit-rates this can do better than JPEG!https://t.co/ATIyOEiwNX
— Emilien Dupont (@emidup) March 10, 2021
with @adam_golinski @notmilad @yeewhye @ArnaudDoucet1 pic.twitter.com/5sVBc2oST5
Want to make your results look "incredible"? Turn a modest improvement into a WIRED story? Step 1: run y axis from 62 to 78 (rather than 0 to 100). Step 2: help yourself to an extra billion parameters. Step 3: congratulate yourself.
— Gary Marcus (@GaryMarcus) March 7, 2021
New SEER paper @facebook pic.twitter.com/y64IXuPy9f
Detectron2Go (D2Go) is a new, state-of-the-art extension for Detectron2 that gives developers an end-to-end pipeline for training and deploying object detection models on mobile devices and hardware.https://t.co/SjtH6PWBQq pic.twitter.com/QeZE4rR74w
— Facebook AI (@facebookai) March 4, 2021