Liyu Zerihun
ልዩ ዘርሁን
Researcher && Research Engineer at RFA Electric · Machine Learning, Physics Simulations, Systems Design
Hello, fellow Homo sapiens.
~/research
3 papers
Invertible Memory Flow NetworksA preliminary study of tree-structured memory for long sequences.
Measuring What Neural Networks Really LearnA framework for estimating neural network intrinsic dimensionality through geometric distillation.
Time Without TimestepsSimulating coupled dynamical systems via self-consistency.
~/projects
6 projects
RitualsApp for tracking and improving daily habits.
EVE: Technical OverviewAn autonomous AI developer that implements, tests, learns from failures, and iterates on real codebases.
NeuroEvolutionEvolutionary approaches to neural systems.
ls --all → 6 projects
~/writing
6 essays
Looking at the MoonOn race, determinism, and identity.
Snowball CriterionA criterion for advanced intelligence: the ability to design a more intelligent successor.
Why We're Alone: The Gradient Descent HypothesisA Great Filter hypothesis on convergent optimization and extinction.
ls --all → 6 essays
~/about
A computer science degree teaches you discrete machines and symbolic algorithms. Deep learning teaches you continuous machines and the algorithms that run on them. Most of my work lives at the boundary between the two: taking discrete or constrained systems and finding ways to make them continuous and differentiable.
© 2026 Liyu Zerihun · ልዩ ዘርሁን
liyulg0@gmail.com · github.com/LiyuZer