Julien Jerphanion writes in-depth articles on machine learning performance, numerical computing, and software craftsmanship. His work focuses on scikit-learn internals, efficient algorithms, and building a strong theoretical understanding of ML systems.
9 articles from this blog
A guide to long-term best practices for managing open-source projects, covering organization, PRs, issues, security, and releases.
Analyzes performance improvements and hardware scalability of the PairwiseDistancesArgKmin algorithm in scikit-learn's k-nearest neighbors implementation.
Introducing PairwiseDistancesReduction, a new Cython-based abstraction in scikit-learn for high-performance CPU computations of reductions over pairwise distances.
Analyzes performance bottlenecks in scikit-learn's k-nearest neighbors search and introduces a new implementation for better CPU scalability.
Analyzes performance limitations in scikit-learn due to CPython internals, memory hierarchy issues, and lack of low-level data structures.
Explains ongoing developer efforts to dramatically improve scikit-learn's performance, focusing on hardware scalability and algorithmic optimizations.
Explains efficient vectorized methods for sampling points from spline curves and 2-sphere splines using linear algebra and caching techniques.
Explores AI-driven content curation through Archillect and algorithmic generative art via Inconvergent, highlighting automated creativity.
A deep dive into designing and implementing a Multilayer Perceptron from scratch, exploring the core concepts of neural network architecture and training.
