← Search

Albert Miao

2 accepted papers

2026

Features Emerge as Discrete States: The First Application of SAEs to 3D Representations

ICLR 2026poster

Sparse Autoencoders (SAEs) are a powerful dictionary learning technique for decomposing neural network activations, translating the hidden state into human ideas with high semantic value despite no external intervention or guidance. However, this technique has rarely been applied outside of the text…

Cited by 0SourceScholar
2026

Quartet of Diffusions: Structure-Aware Point Cloud Generation through Part and Symmetry Guidance

ICLR 2026poster

We introduce the *Quartet of Diffusions*, a structure-aware point cloud generation framework that explicitly models part composition and symmetry. Unlike prior methods that treat shape generation as a holistic process or only support part composition, our approach leverages four coordinated diffusio…

Cited by 0SourceScholar