Depth-Dependent Hidden-State Collapse in Dynamical System Autoencoders for LiDAR Point-Cloud Classification
arXiv:2607.14463
The paper examines Dynamical System Autoencoders for LiDAR point‑cloud classification and discovers that increasing encoder depth to five layers causes a hidden‑state collapse, producing nearly constant representations that limit class separation and classification performance.
Abstract
We study Dynamical System Autoencoders (DSAE) for LiDAR point-cloud classification using spatial coordinates and Product Coefficient feature augmentations. The experiments compare separately trained DSAE architectures at encoder depths and evaluate the resulting hidden representations with Random Forest, kNN, and a majority-class Dummy baseline. The main finding is a hidden-state collapse at . For both xyz and xyz plus Product Coefficient inputs, the hidden-state standard deviation falls to the order of , while all three classifiers attain the same macro F1 score of . We prove that between-class hidden scatter is bounded by total hidden scatter, which in turn is controlled by the reported hidden-state variance. Thus a nearly constant hidden representation cannot retain substantial class-separating structure. Product Coefficients neither improve pre-collapse macro F1 nor prevent the collapse in the present DSAE setting. These results identify large-depth representation collapse as a concrete failure mode for DSAE LiDAR classification.
6 pages, 2 figures, 1 table. Submitted to the 2026 IEEE High Performance Extreme Computing Conference (HPEC 2026)