About

Who we are

We are a small team who did particle-physics analysis before machine learning was standard on it — and then had to pick up the ML the hard way, mostly by getting it wrong in front of a review committee first.

Everything here is written for one reader: a physicist who can do the maths, has never been shown the ML workflow end to end, and does not want to be sold hype. That is the gap we kept hitting ourselves — scattered GitHub notebooks, CS-flavoured tutorials that rename things you already understand, and no one laying out the path in the order you would actually use it.

How we work

The material follows the same order a real analysis does — frame, build, train, apply, check — with the traps flagged at the exact step where people fall into them. No framework worship, no "AI changes everything" talk. The goal is a result you can defend in review.

We publish under the team name rather than individual bylines — the method stands on the citations and the code, not on personalities.

What we are not

We are not a coding bootcamp. Generic machine-learning tutorials chase a leaderboard score and stop. Physics analysis has to survive systematic-uncertainty checks, data-versus-simulation agreement, and a convener who will ask why your result can be trusted. Machine learning earned its place in physics precisely because it handles these problems well (Roe et al., NIM A 543, 2005), and the ideas underneath it are decades old and well understood (Freund & Schapire, J. Comput. Syst. Sci. 55, 1997) — not a passing trend.

We are not selling a black box. We teach the same open, peer-reviewed tools the field actually uses (Hoecker et al., arXiv:physics/0703039, 2007; Chen & Guestrin, KDD 2016) — and we show you how to interrogate them, not just press run.

We are not overstating what machine learning does. Where its advantage is modest or context-dependent, we say so. Benchmarks like the Higgs Machine Learning Challenge (Adam-Bourdarios et al., JMLR W&CP 42, 2015) show these methods are strong but not magic; the win comes from disciplined checking, not from the method alone.

References

  • Roe, B. P., Yang, H.-J., Zhu, J., Liu, Y., Stancu, I., & McGregor, G. (2005). Boosted decision trees as an alternative to artificial neural networks for particle identification. Nuclear Instruments and Methods in Physics Research A, 543(2-3), 577-584.
  • Freund, Y., & Schapire, R. E. (1997). A decision-theoretic generalization of on-line learning and an application to boosting. Journal of Computer and System Sciences, 55(1), 119-139.
  • Hoecker, A., et al. (2007). TMVA — Toolkit for Multivariate Data Analysis. arXiv:physics/0703039.
  • Chen, T., & Guestrin, C. (2016). XGBoost: A Scalable Tree Boosting System. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 785-794.
  • Adam-Bourdarios, C., Cowan, G., Germain, C., Guyon, I., Kegl, B., & Rousseau, D. (2015). The Higgs boson machine learning challenge. JMLR Workshop and Conference Proceedings, 42, 19-55.

Start where we would start

The best way to see how we teach is to read something we made. Get the free Module 1 workbook and the Roadmap — the exact path from where you are now to a result you can defend in review, no payment required. If it is useful, the complete course picks up from there.

Questions? Reach us at [email protected].