Skip to content

DSB #110


그것은금요일 and weekend and bulletin! Last volume was amazing and this one is amazing as well. For me it was interesting to learn about vokenization from Papers & Books or read about meta-loss landscapes in Analytical. But every article is full of interesting info. So go for it.

And as always, enjoy your reading.

Analytical – Have you ever given a thought to exploration of hyperparameter meta-loss landscapes? No? Then now is the time to find out if there are better solutions than the gradient based method. (rcmd by reader) – Breathtaking solutions for Advent of Code 2020 in Python. (rcmd by reader) – Our former colleague wrote a series of articles about portfolio optimization in Python. First two articles are more theoretical, the third one includes code. (rcmd by reader)

Computer Science & Science – Is Haskell and Hasktorch a future of datascience? Hopefully not, or…? (rcmd by reader) – Data Science is not only about modelling, but also about data engineering. You train it, you run it! Why is that so important, can be found also here. (rcmd by reader) – Epic introduced their MetaHuman Creator where you can in less than hour create a digital human of pixel perfect quality.

Graphs and Visualizations – Graphs neural networks should be noticed – the link will tell you how to apply them in practice. (rcmd by reader) – Kolmogorov complexity and squishy bunnies explained by math and nice vizualizations. – Simulation of tectonic plates movement.

Business and Career – Data scientists are still in demmand and still there is a shortage of them. Why is that and how to handle it is answered in the article. – Robinhood showed that fintechs are powerfull and maybe, (un)fortunately, we may expect some regulations. Not only because of the stock market but also because of shadow banking and more. – I love it, this article shows differences in thinking between big finance institutions like banks and fintechs. Put it shortly, to have mobile apps is not enough.

Pop – Why big companies like Tesla invest into Bitcoin? Why, why, why? Yes, there are reasons. – I really like this parable of ML to Library of Babel – ML enhances our world, creates more possibilities, but where does it lead?–how-ml-saves-us–1-7m-a-year-on-document-previews – Dropbox saved 2 mil. USD thanks to optimizing documents preview.

Education – Match statement for pattern matching is new for Python 3.10. The quickest intro is in Appendix A. (rcmd by reader) – Linear transformation of 2D space using matrices, simply and intuitively with great visualizations + code in Python. (rcmd by reader) – This is gold. Reddit thread about deep learning theory with many interesting sources.

Data & Libraries –  Generate fake TLS certs to use in your tests with Python library trustme. (rcmd by reader) – qlib is a python library by Microsoft for AI quantitative investing. – Quickly deploy your models with BudgetML – seems interesting for particular utilization when you don’t care about details.

Video & Podcast – Podcast about containers – Kubernetes, Heroku, Digital Ocean and many more. (rcmd by reader) – rstudio::global 2021 is a conference that was held in January and lectures are already online. In one of them, for example, Michael Chow transformed Tidyverse to Python. (rcmd by reader) – Is it possible to be so lucky, that it is virtually impossible? (rcmd by reader)

Papers & Books – Learn these new terms: vokenization and vokens (visual tokens). You convert the language tokens to related images and achieve a visual supervision of the language model. Code is here. (rcmd by reader) – Graph embedding of commodities for recommendation system in Alibaba. (rcmd by reader) – One serious optimization of GPU computation with help from CPU. It’s almost unbelievable. With one GPU you can achieve the same speed as with multiple GPUs before. Code is here.

Behind the Fence – Data scientist in Air-Conditioning, Heating, and Refrigeration Institute, Arlington, Virginia, USA.

Joke – Oh my 😀 Scroll down to see others, they are hilarious.

One Comment

Leave a Reply