today is Sunday which is unusual for DSB, but at least you have something pleasurable to do today or on Monday. I really recommend the first two articles about meta-overfitting or branch specialization from Analytical.
As always, enjoy your reading.
http://www.offconvex.org/2021/04/07/ripvanwinkle/ – Meta-overfitting error (MOE) and its upper bounds for popular deep net architectures. The estimate is called Rip van Winkle’s Razor.
https://distill.pub/2020/circuits/branch-specialization/ – The branch specialization – when in neural networks across different architectures and tasks branches specilize on certain problems. The features are organizing between branches.
https://rachitsingh.com/deep-learning-model-compression/ – State-of-the-art methods for neural networks optimization.
Computer Science & Science
https://pythonspeed.com/articles/pandas-sql-chunking/ – Pandas and batching means unlimited memory for everybody. More articles on this topic can be found here. (rcmd by reader)
https://towardsdatascience.com/compositional-ai-the-future-of-enterprise-ai-3d5289dfa888 – Usualy companies build single ML solution, but compositional AI is the next step when you combine what you already have to create something new. (rcmd by reader)
https://www.emilwallner.com/p/ml-rig – What a beauty, 4 NVIDIA RTX A6000, AMD EPYC 2 with 32 cores, 192 GB in GPU memory and 256 GB in RAM…
Graphs and Visualizations
https://pair-code.github.io/lit/demos/ – Ever wonder how to visualize NLP models? Then stop wondering and learn. (rcmd by reader)
https://ivelasq.rbind.io/blog/other-geoms/ – Some atypical but very appealing ggplot geoms.
Business and Career
– Is KBC leaving Ireland? What does it mean for customers? And some interesting Q&A. (rcmd by reader)
https://www.forbes.com/sites/ronshevlin/2021/04/12/embedded-fintech-versus-embedded-finance-jumpstarting-new-product-innovation-in-banks/ – Embedded finance and embedded fintechs and their role in banking.
https://www.nytimes.com/2021/04/15/business/coinbase-market-debut.html – Coinbase is now, since Wednesday, a publicly traded company and it’s quite huge. There is a list of Coinbase acquisitions and interesting comparison with Amazon.
https://towardsdatascience.com/towards-the-end-of-deep-learning-and-the-beginning-of-agi-d214d222c4cb – Static vs dynamic paradigm in deep learning architectures and the role of adversarial examples. (rcmd by reader)
https://www.bbc.com/news/technology-56745730 – Facial recognition and some wild algorithms will probably be banned by the EU. They are already people falsely accused and arrested thanks to these technologies. Also it is a topic of a new documentary by Netflix Coded Bias. (rcmd by reader)
https://mattturck.com/madindex/ – You probably know S&P500 and you should also know the MAD index that follows in US publicly listed ML/AI companies.
https://ai.googleblog.com/2021/03/recursive-classification-replacing.html – Google AI comes with an example of success as a new way how to teach an agent in RL. (rcmd by reader)
https://www.analyticsvidhya.com/blog/2021/04/a-guide-to-feature-engineering-in-nlp/ – Some basic ways how to create features in NLP.
https://gretteljuarez.medium.com/using-googles-speech-to-text-api-with-python-e436dc6cb944 – Google speech2text api in Python, how to do it and what does it require.
Data & Libraries
https://blog.google/technology/area-120/stack/ – Stack is a new experimental app by Google that digitizes your physical receipts, which is the domain of many startups. Now the big guy is coming… (rcmd by reader)
https://www.analyticsvidhya.com/blog/2021/04/the-easiest-way-to-deploy-machine-learning-models-pywebio/ – How to use PyWebIO to deploy a model.
https://towardsdatascience.com/how-to-create-mathematical-animations-like-3blue1brown-using-python-f571fb9da3d1 – 3Blue1Brown has propably the best videos about math on the youtube, but he also have great cool animation and you can have too! With Manim and Python!
Video & Podcast
https://open.spotify.com/show/4QodYB2UKaa20YgT5j387D?si=ZlqVXWjCQ72c7bTwlvp84w&nd=1 – Podcast Banking Transformed with Jim Marous who is debating the future of financial institutions in digital world. Lot’s of episodes to listen to. (rcmd by reader)
https://www.youtube.com/watch?v=U4ogK0MIzqk – Nicely done video about creating a chess program in C#. (rcmd by reader)
Papers & Books
https://arxiv.org/abs/2103.14749 – The most used datasets for computer vision and nlp have label errors. And it’s mainly important for model evaluation on test sets.
Behind the Fence
https://jobs.lever.co/atomic/a00c38dc-4a8f-4131-a00a-a754cd99218a – Lead Data Scientist in Butter Payments, San Francisco, US.
https://ibb.co/593YzzT – Surprise! (rcmd by reader)