| | At Fidelity, every decision is grounded in strategy: why use this technology? Is it practical and useful? How do we build it, and how do we make it better? Will it help improve our clients' lives? We make room for our technologists to ask these questions and to learn through real-world practice, organized roadmaps, and clear direction. Read more about how Fidelity's approach to tech can shape your career.
*Message from this week's sponsor, Fidelity Careers |
| | |
| | 🔗 Note: if you don’t see links in this email, it’s also available on the web: dataelixir.com/archive/issue-584.html |
| | |
| | Some famous Erdős problems are starting to fall to AI, but the real story here is the workflow. Researchers are combining generation, critique, and verification in ways that show how AI may tackle other hard technical problems too.
Quanta Magazine | Konstantin Kakaes |
| | |
| | | Here’s what applied math looks like when the output is a painting. Gaussian splats, eigenvectors, wavelets, noise, and GPU rendering are layered together step by step, with visual examples showing why each technique matters.
Dmitri Sotnikov |
| | |
| | This is a good case study in building just enough NLP for a constrained problem. Start with simple word matching, see where it breaks, then add stemming, bigrams, fuzzy matching, and a few exceptions.
Ivan Sagalaev |
| | |
| | Golf is the vehicle here, but the interesting part is the modeling problem. Matthew Schoolfield walks through how he built a synthetic dataset from limited data, domain knowledge, and statistical modeling, then used it to create a realistic golf shot dispersion simulator.
Matthew Schoolfield |
| | |
| | A model can be mathematically sound and still get a shrug from the person who needs to use it. This post starts there, then works backward through customer lifetime value modeling and the creation of a lightweight Python replacement for the now-archived lifetimes library.
clv[kit] | Nasser Boan |
| | |
| | Parallelism can change the order of floating-point operations, which means numerical results aren’t always as stable as you might think. This post walks through a replication failure using Stata, but the underlying issue applies across numerical computing tools.
Office of the AEA Data Editor |
| | |
| | | | GeoLibre is a full-featured open-source GIS platform for visualizing, exploring, and analyzing geospatial data. It runs in a web browser and your data stays local. This project has only been public for a few months, but there’s already a lot here and it has over 5000 stars on GitHub.
GitHub | Open Geospatial Solutions |
| | |
| | | The Pudding asked thousands of people to mow virtual lawns, then compared their routes with optimal solutions. It turns out the best problem solvers don't necessarily think longer. They think at the right moments.
The Pudding | Russell Samora, Florina Sutanto, Arjun Kakkar, and Sushant Karki |
| | |
| | | This week's issue sponsored by: | | |
|
