Top 8 trends from ICLR 2019

Overview of trends on #ICLR2019:
1. Inclusivity
2. Unsupervised representation learning & transfer learning
3. Retro ML
4. RNN is losing its luster with researchers
5. GANs are still going on strong
6. The lack of biologically inspired deep learning
7. Reinforcement learning is still the most popular topic by submissions
8. Most accepted papers will be quickly forgotten

Link: https://huyenchip.com/2019/05/12/top-8-trends-from-iclr-2019.html

#ICLR #overview

​​The new ResNet PoseNet model is much more accurate than the MobileNet one (the trade off being size & speed). The model is quantized & 25MB.
Pose estimation model, capable of running on devices

This model is really great for art installations or running on desktops.

Demo (requires camera, will work on desktop): https://storage.googleapis.com/tfjs-models/demos/posenet/camera.html?linkId=69346544
Github: https://github.com/tensorflow/tfjs-models/tree/master/posenet

#tensorflow #tensorflowjs #js #pose #poseestimation #posenet #ResNet #device #ondevice

Estimating the success of re-identifications in incomplete datasets using generative models

99.98% of Americans would be correctly re-identified in any dataset using 15 demographic attributes, suggesting that even heavily sampled anonymized datasets are unlikely to satisfy the modern standards for anonymization set forth by GDPR.

This is a big concern about privacy and a problem for Data Engineering, especially for those working with anonymized personal information. Paper provides a way to re-identify person from anonymized dataset, this can be useful for people who work for government or security companies

https://www.reddit.com/r/science/comments/chko43/9998_of_americans_would_be_correctly_reidentified/

#privacy #gdpr #federatedlearning #ml

🔥Neural Point-Based Graphics

Breakthrough work on generating realistic 3D scenes in AR.

Demo: https://www.youtube.com/watch?v=7s3BYGok7wU
Project page: https://dmitryulyanov.github.io/neural_point_based_graphics
ArXiV: https://arxiv.org/pdf/1906.08240.pdf

#CV #AR #DL #SamsungAI

​​Baidu’s Optimized ERNIE Achieves State-of-the-Art Results in Natural Language Processing Tasks

#Baide developed ERNIE 2.0, a continual pre-training framework for language understanding. The model built on this framework has outperformed #BERT and #XLNet on 16 tasks in Chinese and English.

Link: http://research.baidu.com/Blog/index-view?id=121

#NLP #NLU

​​🔥Interactive demo of GAN turning doodles into beautiful pictures

NVidia released #GauGAN for anyone to use. Trained on 1M images, the #GAN tool automatically turns doodles into photorealistic landscapes.

Project page: https://www.nvidia.com/en-us/research/ai-playground/
Interactive demo: http://nvidia-research-mingyuliu.com/gaugan

#Nvidia #CV #DL

Great article on text preprocessing, covering cleaning, #tokenization, #lemmatization and other aspects

Link: https://medium.com/@datamonsters/text-preprocessing-in-python-steps-tools-and-examples-bf025f872908

#NLP #NLU #datacleaning

21k followers — best feedback from the audience!

Thank you!

Facebook open sourced video alignment algorithms that detect identical and near identical videos to build more robust defenses against harmful visual content.

Project page: https://newsroom.fb.com/news/2019/08/open-source-photo-video-matching/
Code: https://github.com/facebookresearch/videoalignment

#Facebook #video #cv #dl

ODS breakfast in Paris! See you this Saturday at 10:30 at Malongo Café, 50 Rue Saint-André des Arts.

​​New paper on training with pseudo-labels for semantic segmentation

Semi-Supervised Segmentation of Salt Bodies in Seismic Images:
SOTA (1st place) at TGS Salt Identification Challenge.

Github: https://github.com/ybabakhin/kaggle_salt_bes_phalanx
ArXiV: https://arxiv.org/abs/1904.04445

#GCPR2019 #Segmentation #CV

​​Long-form question answering

Facebook AI shared the first large-scale data set, code, and baseline models for long-form QA, which requires machines to provide long, complex answers — something that existing algorithms have not been challenged to do before.

Link: https://ai.facebook.com/blog/longform-qa/

#FacebookAI #Facebook #NLP #NLU #QA

PyTorch for research

PyTorch Lightning — The PyTorch Keras for ML researchers. More control. Less boilerplate.

Github: https://github.com/williamFalcon/pytorch-lightning

#PyTorch #Research #OpenSource

spaCy meets PyTorch-Transformers: Fine-tune BERT, XLNet and GPT-2

Including pretrained models.

Link: https://explosion.ai/blog/spacy-pytorch-transformers
Pip: pip install spacy-pytorch-transformers

#Transformers #SpaCy #NLP #NLU #PyTorch #Bert #XLNet #GPT

ODS breakfast in Paris! See you this Saturday at 10:30 at Malongo Café, 50 Rue Saint-André des Arts.

You do not need any presentation or preparation, this is free format for people who are around these days and wanna chat with fellow data scientists

​​Photo to anime portrait

U-GAT-IT — Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translation.

Link: https://github.com/taki0112/UGATIT

#Tensorflow #GAN #CV #DL #anime

​​Unified rational protein engineering with sequence-only deep representation learning

UniRep predicts amino-acid sequences that form stable bonds. In industry, that’s vital for determining the production yields, reaction rates, and shelf life of protein-based products.

Link: https://www.biorxiv.org/content/10.1101/589333v1.full

#biolearning #rnn #Harvard #sequence #protein

A Guide for Making Black Box Models Explainable.

One of the biggest challenges is to make ML models interpretable (explainable to human, preferably, non-expert). It matters not only in terms of credit scoring, to exclude possibility of racism or any other bias or news promotion and display (Cambridge Analytica case), but even in terms of debug and further progress in model training.


Link: https://christophm.github.io/interpretable-ml-book/

#guide #interpretablelearning #IL

🔥 New Releases: PyTorch 1.2, torchtext 0.4, torchaudio 0.3, and torchvision 0.4

https://pytorch.org/blog/pytorch-1.2-and-domain-api-release/

https://github.com/pytorch/pytorch/releases