Links of the week #39

Self-driving cars

• The Waymo Way: Making Autonomous Driving a Reality (Sep. 2024), with Dmitri Dolgov, co-CEO of Waymo

• But before reaching true L4 or L5, the fact is that Partial Autonomous Driving Leads to Distracting Behavior (Sep. 2024), via Tom Dahlstr?m

“These results are a good reminder of the way people learn,” said IIHS President David Harkey. “If you train them to think that paying attention means nudging the steering wheel every few seconds, then that’s exactly what they’ll do.”

Physics (+ Generative AI)

• What is Entropy? by the american physicist John Baez (I remember reading him 25 years ago on Usenet group sci.physics :) ) This paper (also available on his academic page) is being used by people experimenting with Google's NotebookLM "podcast feature". I suppose this is the case of this version : https://www.youtube.com/watch?v=1LTHoKy2bFw.

Once there was a thing called Twitter, where people exchanged short messages called ‘tweets’. While it had its flaws, I came to like it and eventually decided to teach a short course on entropy in the form of tweets. This little book is a slightly expanded version of that course.

It’s easy to wax poetic about entropy, but what is it? I claim it’s the amount of information we don’t know about a situation, which in principle we could learn. But how can we make this idea precise and quantitative? To focus the discussion I decided to tackle a specific puzzle: why does hydrogen gas at room temperature and pressure have an entropy corresponding to about 23 unknown bits of information per molecule?

Generative AI + coding

• GenAI Assisted Coding with Cursor and Google Colab (Sep. 2024) by John Hedengren . A really simple introduction, useful for people who had never used these tools (not only Colab, but rather Gemini with Colab)

Model reduction

• Beyond POD: The Power of Spectral Proper Orthogonal Decomposition (Sep. 2024) by Shubham Goswami, PhD

POD often falls short when dealing with systems exhibiting rich temporal dynamics. This is where Spectral Proper Orthogonal Decomposition (SPOD) shines. By incorporating frequency information into the analysis, SPOD unveils a deeper understanding of the underlying mechanisms driving system behavior

CAD

• Dassault’s CAD/CAM | From Concept to Reality : "Digitized version of a 16mm film reel. Showing the usage of Dassault’s CAD/CAM in the early 80’s"

Python

• Accelerate Pandas 20x using FireDucks (Sep. 2024), by Avi Chawla : "As per FireDucks’ official benchmarks, it can be ~20x faster than Pandas and ~2x faster than Polars, as shown below on several queries". An alternative to Polars to improve Pandas efficiency ?

Julia

• Julia for Engineers: Part 1 Algorithms (Sep. 2024)

Optimisation

• Scaling up linear programming with PDLP (Sep. 2024) by Google Research

Utilizing matrix-vector multiplication rather than matrix factorization, PDLP requires less memory and is more compatible with modern computational technologies like GPUs and distributed systems, offering a scalable alternative that mitigates the memory and computational inefficiencies of traditional LP methods. PDLP is open-sourced in Google’s OR-Tools. This project has been in development since 2018 (...), and we are proud to announce that it was co-awarded the prestigious Beale — Orchard-Hays Prize at the International Symposium of Mathematical Programming in July 2024. This accolade is one of the highest honors in the field of computational optimization, awarded every three years by the Mathematical Optimization Society.