2020 DS/ML digest 12

2020 DS/ML digest 12

Posted by snakers41 on October 26, 2020

Internet / tech

The end of American Internet https://www.ben-evans.com/benedictevans/2020/10/3/the-end-of-the-american-internet
If you had any doubts that US is a police state - https://thegradient.pub/how-the-police-use-ai-to-track-and-identify-you/
CORPORATE VENTURE CAPITAL https://digitstodollars.com/2020/10/15/corporate-venture-capital/

Tools

Neural network visualization - https://github.com/lutzroeder/netron
Containerization landscape - https://martinheinz.dev/blog/35
New Python release - https://towardsdatascience.com/python-3-9-9c2ce1332eb4
Context managers in Python - https://martinheinz.dev/blog/34

Research

Advancing NLP with Efficient Projection-Based Model Architectures

If you still have some illusions about OpenAI https://thegradient.pub/ai-democratization-in-the-era-of-gpt-3/
Transformers are graph neural networks https://thegradient.pub/transformers-are-graph-neural-networks/
Finally someone understands dual-licensing and GPL - https://blog.cerebralab.com/Dual licensing GPL for fame and profit
Launch costs to low Earth orbit, 1980-2100 - https://www.futuretimeline.net/data-trends/6.htm
New wave of space-tech - https://digitstodollars.com/2020/10/06/consumers-the-final-frontier-of-space/, http://digitstodollars.com/2020/10/08/the-buck-in-buck-rodgers/
The Gap: Where Machine Learning Education Falls Short - https://thegradient.pub/the-gap-where-machine-learning-education-falls-short/
A Quarter Century of Hype - 25 Years of the Gartner Hype Cycle https://vimeo.com/464835556 - interesting that speech is listed as mature here
40 milliseconds of latency that just would not go away https://rachelbythebay.com/w/2020/10/14/lag/
It’s complicated. A deep dive into the Viz/Medicare AI reimbursement model - https://lukeoakdenrayner.wordpress.com/2020/09/24/its-complicated-a-deep-dive-into-the-viz-medicare-ai-reimbursement-model/
Recreating Historical Streetscapes Using Deep Learning and Crowdsourcing - https://ai.googleblog.com/2020/10/recreating-historical-streetscapes.html
ODS paper digest - https://habr.com/ru/company/ods/blog/523268
HD face swap - https://studios.disneyresearch.com/2020/06/29/high-resolution-neural-face-swapping-for-visual-effects/
ML filters in Photoshop - https://blogs.nvidia.com/blog/2020/10/20/adobe-max-ai/
Russian large GPT by Sber - https://habr.com/ru/company/sberbank/blog/524522/
Yet another fruitless Russian corpus repackaging - https://omnia-russica.github.io/
Data distullation - https://dyakonov.org/2020/10/21/data-distillation/
Rethinking Attention with Performers - https://ai.googleblog.com/2020/10/rethinking-attention-with-performers.html
https://ai.googleblog.com/2020/09/improving-sparse-training-with-rigl.html

  • Looks like compact networks are indeed longer to train

Hardware

RTX3090 review https://www.pugetsystems.com/labs/hpc/RTX3090-TensorFlow-NAMD-and-HPCG-Performance-on-Linux-Preliminary-1902
Lenovo officially supports Ubuntu now - https://habr.com/ru/company/lenovo/blog/520678/ ?
https://digitstodollars.com/2020/10/21/semiconductors-midlife-crisis/

Misc

Questions About Trees - http://bit-player.org/2020/questions-about-trees
Simple diagrams in Python for docs

Datasets

Large radiology dataset - https://portal.imaging.datacommons.cancer.gov/

  • 3,230 cases
  • 1.03 TB Data Volume
  • 49,818 Image Series