Deconstructing Neural ODEs from Scratch
Explore the mathematics of continuous-depth networks, ODE solvers, and infinite-layer architectures — built entirely from scratch in pure PyTorch.
Part 1
The Math of Continuous Depth
Breaking down ODE initial value problems, Euler and RK4 solvers, the adjoint method, and the continuum limit of ResNets.
Part 2
PyTorch Implementation
Building ODE solvers, Neural ODE layers, and continuous-depth classifiers — entirely from scratch.
View Code on GitHub
Part 3
Infinite Layers, Finite Memory
Spiral classification with 19.3x fewer parameters than a discrete ResNet, comparing Euler vs RK4, and visualizing learned flow fields.