Top all-time · #7838
Francis Bach
Personal blog · francisbach.comaimath
Francis Bach writes Machine Learning Research Blog, a personal blog at francisbach.com about ai and math. Since 2019, 10 different Hacker News readers have linked it 20 times, most often “Learning Theory from First Principles”. It is still writing: the last post was on 2026-09-25, and you can follow it by feed.
Too few sampled comments (3) to say how often HN recommends it.?
Why HN links it
HN readers link Francis Bach's posts on SGD noise and his book Learning Theory from First Principles. The commenters cite them as research background and as a sourced explanation of learning theory.
Summarised by Claude from a sample of 3 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Most-linked posts?
- Learning Theory from First Principles3 people · 3 mentions · first linked 2025-01-06
- Rethinking SGD’s Noise2 people · 2 mentions · first linked 2022-07-25
- Polynomial magic I: Chebyshev polynomials2 people · 2 mentions · first linked 2019-11-04
- On the unreasonable effectiveness of Richardson extrapolation2 people · 2 mentions · first linked 2020-03-01
What Hacker News says?
In general you do, because the unbiased estimates have higher generalization error. You are already dealing with sampling noise. I am not an expert in optimization, and what "poorly understood" means to you, but I know there is quite some research on the properties of SGD noise; e.g., https://francisbach.com/rethinking-sgd-noise/ Dissecting the Effects of SGD Noise in Distinct Regimes of Deep Learning https://arxiv.…
Comment linking /rethinking-sgd-noise/
on “How linear regression works intuitively and how it leads to gradient descent”Submitted Revisiting the Classics: Jensen's Inequality (2023) as a story · 89 points
HN discussionBut why does, as you explain "training goes brrr"? Francis Bach, the author, makes a good faith effort to explain exactly why this material is beneficial (see https://francisbach.com/my-book-is-out/ ): "Why yet another book on learning theory? ...the main reason is that I felt that the current trend in the mathematical analysis of machine learning was leading to overly complicated arguments and results that are ofte…
Comment linking /my-book-is-out/
on “Learning Theory from First Principles [pdf]”Submitted Exploding variance of means of exponentials: least-squares to the rescue as a story · 65 points
HN discussionhttps://francisbach.com/my-book-is-out/ https://mitpress.mit.edu/9780262049443/learning-theory-from-...
Comment linking /my-book-is-out/
on “Learning Theory from First Principles [pdf]”Submitted Scaling Laws of Optimization as a story · 11 points
HN discussionSubmitted Computer-Aided Analyses in Optimization as a story · 5 points
HN discussionSubmitted On the unreasonable effectiveness of Richardson extrapolation as a story · 5 points
HN discussionSubmitted Learning Theory from First Principles as a story · 3 points
HN discussionSubmitted Polynomial magic I: Chebyshev polynomials as a story · 3 points
HN discussionSubmitted Beyond Power Laws: Scaling Laws for Next-Token Prediction as a story · 2 points
HN discussionSubmitted Polynomial magic I: Chebyshev polynomials as a story · 2 points
HN discussionSubmitted Polynomial Magic II: Jacobi Polynomials as a story · 2 points
HN discussionSubmitted The unreasonable effectiveness of Richardson extrapolation as a story · 1 points
HN discussionSubmitted Unraveling spectral properties of kernel matrices as a story · 1 points
HN discussionSubmitted Revisiting the Classics: Jensen’s Inequality as a story · 1 points
HN discussionSubmitted Rethinking SGD’s Noise as a story · 1 points
HN discussionSubmitted The many faces of integration by parts – I: Abel transformation as a story · 1 points
HN discussionSubmitted Gradient descent for wide two-layer neural networks – I: Global convergence as a story · 1 points
HN discussionSubmitted The sum of a geometric series is all you need as a story · 1 points
HN discussion
