New Julia Lecture: Differentiable Filters and Enzyme Updates
9 Feb 2026
A new lecture has been added to Quantitative Economics with Julia: Differentiable Filters.
This lecture covers implementing and differentiating the Kalman filter using both forward-mode and reverse-mode automatic differentiation. It documents specific coding patterns needed to get high performance while remaining compatible with Enzyme.
A key challenge addressed in the lecture is that small, static immutable matrices call for a completely functional coding style, while large matrices require everything to be done in-place. Getting the same code to work efficiently in both cases requires careful design. Much of this is driven by Enzyme’s requirement for non-allocating code — which, fortunately, is usually aligned with highest performance anyway.
This release also updates the lectures to support the official Enzyme.jl release for Julia 1.12.
These updates were developed by Jesse Perla.
Published by: QuantEcon