Data Science Digest

2021-06-04 17:00:06 #DataScienceMemes Open / Comment

2021-06-04 10:00:13 CogView: Mastering Text-to-Image Generation via Transformers

Text-to-Image generation is a challenging task that requires powerful generative models and cross-modal understanding. CogView is a 4-billion-parameter Transformer with VQ-VAE tokenizer that, according to the authors, achieves a new state-of-the-art FID on blurred MS COCO, outperforms previous GAN-based models and a recent similar work DALL-E.

Paper — https://bit.ly/3chno8b
Code — https://bit.ly/3ciDoqp
Demo — https://bit.ly/3z4Ba7N Open / Comment

2021-06-03 13:14:15 Fraud Detection: Using Relational Graph Learning to Detect Collusion

Uber’s popularity attracted the attention of financial criminals in cyberspace. One type of fraudulent behavior is collusion, a cooperative fraud action among users. In this article, Uber Engineering demonstrates a case study of applying a cutting-edge, deep graph learning model called relational graph convolutional networks (RGCN) to detect such collusion.

https://ubr.to/3irmc6f Open / Comment

2021-06-02 22:49:45 DataScience Digest pinned a photo Open / Comment

2021-06-02 22:37:20 ​​Data Science Digest — 02.06.21

The new issue of DataScienceDigest is here! Hop to learn about the latest news, articles, tutorials, research papers, datasets, videos, and tools on DataScience, AI, ML, and BigData. All sections are prioritized for your convenience. Enjoy!

https://bit.ly/3vN2CF4

Join @DataScienceDigest Open / Comment

2021-06-02 19:03:48 Albumentations 1.0.0 has been released!

Albumentations is a computer vision tool and a Python library designed to improve the performance of deep convolutional neural networks by enabling fast, flexible, cost- and resource-efficient image augmentations. The tool can be used for different CV tasks, including object classification, segmentation, and detection.
New version contains 10 new transforms, independence from imgaug, bug fixes, etc.

https://bit.ly/3fKM6jC Open / Comment

2021-06-01 09:30:12 Build a Scalable Machine Learning Pipeline for Ultra-High Resolution Medical Images using Amazon SageMaker

In this comprehensive article by the AWS team, you’ll learn how to preprocess medical images in ultra-high resolution, train an image classifier on these preprocessed images, and deploy a pretrained model as an API — all done on the Amazon SageMaker platform — to, finally, build a highly scalable machine learning pipeline.

https://amzn.to/3fAJowT Open / Comment

2021-05-31 09:30:17 GAN Prior Embedded Network for Blind Face Restoration in the Wild

In this paper, Tao Yang et al. use existing generative adversarial network-based methods to solve the problem of blind face restoration from severely degraded face images in the wild. The proposed GAN prior embedded network (GPEN) generates visually photo-realistic results, which are significantly superior to BFR methods both quantitatively and qualitatively.

Paper — https://bit.ly/3fCV8PA
Code — https://bit.ly/2SIt04e Open / Comment

2021-05-30 10:00:11 LAPAR: Linearly-Assembled Pixel-Adaptive Regression Network for Single Image Super-Resolution and Beyond

In this paper, the team of researchers propose a linearly-assembled pixel-adaptive regression network (LAPAR), designed and built to deal with a fundamental problem of upsampling a low-resolution (LR) image to its high-resolution (HR) version. LAPAR is highly lightweight and easy to optimize, and helps achieve superb results on SISR benchmarks.

Paper — https://bit.ly/3yYFzcC
Code — https://bit.ly/2RPlPaG Open / Comment

2021-05-29 10:00:11 Image Cropping on Twitter: Fairness Metrics, their Limitations, and the Importance of Representation, Design, and Agency

In this paper, researchers look into fairness and bias issues in Twitter’s automated image cropping system. They found systematic disparities in cropping, identified contributing factors, and to resolve the problem proposed the removal of saliency-based cropping in favor of a solution that better preserves user agency.

Paper — https://bit.ly/3yM4ksa
Code —https://bit.ly/3fSifUX Open / Comment