https://thetrigun99.github.io/A minimal, responsive and feature-rich Jekyll theme for technical writing. 2026-06-12T01:38:38+00:00 Damien Aupretre https://thetrigun99.github.io/ Jekyll © 2026 Damien Aupretre /assets/img/favicons/favicon.ico /assets/img/favicons/favicon-96x96.png 2026-05-18T00:00:00+00:00 2026-06-12T01:38:00+00:00 https://thetrigun99.github.io/posts/DDPM/ Damien Aupretre

Preamble You can find the code linked to this post on my github repo. All of this is written by human, but AI helps me correcting mistakes and improving my markdown coloring skills. Let’s augment the level of difficulty of this blog with some generativ modeling that will be a key part of my future work. More precisely, we’ll focus on diffusion algorithms by using first probabilistic view (DDP...

2026-03-04T00:00:00+00:00 2026-03-14T06:28:48+00:00 https://thetrigun99.github.io/posts/SOTA_lasso/ Damien Aupretre

Introduction The classical LASSO solver based on coordinate descent quickly becomes expensive when the number of variables is much larger than the number of observations ($p≫n$). Modern implementations speed up this solver using screening rules, which safely eliminate variables whose optimal coefficient is zero. In this post, I implement and explain the paper Mind the Duality Gap: Safer Rules ...

2026-01-03T00:00:00+00:00 2026-01-03T00:00:00+00:00 https://thetrigun99.github.io/posts/lasso/ Damien Aupretre

repo github lasso Introduction After studying Ridge in depth, let’s now look at its alter ego: Lasso, or Least Absolute Shrinkage and Selection Operator. We saw that Ridge pulled coefficients toward 0 but never actually zeroed them out. Lasso, on the other hand, can eliminate variables entirely, in other words, it selects variables and keeps only the significant ones. Lasso is a regularizatio...

2025-12-19T00:00:00+00:00 2026-03-14T05:43:37+00:00 https://thetrigun99.github.io/posts/ridge/ Damien Aupretre

Repo Github Ridge Introduction Let’s start from the beginning: OLS regression (Ordinary Least Squares), the standard linear regression and a classic in any statistics course. As a reminder, the closed-form estimator is $\hat{\beta} = (X^{\top} X)^{-1} X^{\top} y$ $X \in \mathbb{R}^{n \times p}$ is the matrix of explanatory variables (design matrix), $y \in \mathbb{R}^{n}$ is the observation ...

2025-11-15T00:00:00+00:00 2025-11-15T00:00:00+00:00 https://thetrigun99.github.io/posts/word2vec/ Damien Aupretre

Introduction When I wanted to build a recommendation algorithm, I came across Word2Vec and more specifically the SGNS (Skip-Gram Negative Sampling) variant. Instead of simply using a ready-made model (like gensim) that I barely understood, I thought: why not recode everything from scratch to truly grasp what’s going on under the hood. The original Word2Vec consists of word embedding to captur...