Recent Posts
Loading Hilberts 2025-04-20 Recently I thought: why are loading spinners such boring shapes, why not fill the loading space, surely CSS is capable of this? Thus was born the Loading Hilbert. Here we show a simple iterate of the Hilbert curve in a square. It is easy to imagine how we could do this for any curve, and even fill spaces dynamically with javascript if we were so inclined (this would…
Bugbash Conference: Software Reliability 2025-04-06 Recently I attended the inaugural BugBash conference, hosted by Antithesis, a small 200 person meetup in DC about software reliability and all that encompasses. The talks do not seem to have been posted to the internet at the moment, and while many of them were good, I don’t think any stood out as something to revisit. Instead of a place to…
Large Lambda Model 2025-02-15 Over the last week I decided to write the inference code for GPT-2 after a many year hiatus from Neural Networks. Depending on what primitives you start from, say if you wrote this with JAX or PyTorch, this is quite straight forward, otherwise it is somewhat less so. After lamenting the lack of perfect tensor library in Haskell, I wrote this directly on top of the…
Functional Chebyshev Approximation 2025-02-01 In a previous post I outlined the Fast Fourier Transform, which has many obvious uses regarding waves and periodic functions. Personally, I have found more use in non-periodic function approximation, which we do largely with polynomials of various orthogonal bases. In an incredible twist of fate, it turns out that the most ubiquitous polynomial basis,…
Trace Estimation 2025-01-11 The other day on twitter, I got involved in a discussion about trace estimation, and decided to write down some thoughts on the matter. Typically this is posed as follows. You are given a matrix A ∈ F N × N A \in F^{N\times N} such that we can only access it through matrix-vector multiplication x ↦ A x x \mapsto Ax . Estimate the trace of the matrix. tr A = ∑ i = 1 N A…
FFT in Haskell and Futhark 2024-12-24 (updated: 2024-12-24 11:11) The Fourier transform is one of the fundamental tools in analysis. From the perspective of approximation theory, it gives us one of the orthogonal function bases, the natural basis for periodic functions. To use this numerically, as with any basis, we must sample from the function to be approximated, and periodic functions have a…
Moving from Hugo to Hakyll 2024-11-09 (updated: 2024-11-09 10:59) Five years ago I decided to rewrite this website with a static site generator, and chose Hugo on a whim. After letting it sit around unused for far too long, I moved it to Hakyll . Now that I’m getting back into Haskell, it was a perfect small project, and Pandoc is unbeatable as a document conversion backend. The built in code…
Cross-Validation Methodology in Materials Science 2018-09-01 Cross-validation is a critical part of statistical methodology, for ad hoc models cross-validation may be the only indication of model performance, and without a reasonable cross-validation methodology serious over-fitting can go undetected. This issue is particularly relevant to domains where small data-sets with a comparatively large…
Learning Hyper-Parameters with Bayesian Optimization 2018-05-01 It is well known in the Machine Learning literature that a Grid Search for optimizing parameters is sub-optimal, and generally outperformed by a Random Search. This project explored using Bayesian Optimization as a sequential search strategy, testing the method on the hyper-parameters of a relatively simple Neural Network performing…