Liu Ziyin

A universal compression theory for lottery ticket hypothesis and neural scaling laws

When Reasoning Meets Its Laws

Ubiquity of Hebbian Dynamics in Complex Learning Rules

Topological Invariance and Breakdown in Learning

On Biologically Plausible Learning in Continuous Time

Neural Thermodynamics: Entropic Forces in Deep and Universal Representation Learning

Formation of Representations in Neural Networks

Three Mechanisms of Feature Learning in a Linear Network

Understanding the Emergence of Multimodal Representation Alignment

Remove Symmetries to Control Model Expressivity and Improve Optimization

Compositional Generalization via Forced Rendering of Disentangled Latents

Type-II Saddles and Probabilistic Stability of Stochastic Gradient Descent

Noise Balance and Stationary Distribution of Stochastic Gradient Descent