HH-SAE: Discovering and Steering Hierarchical Knowledge of Complex Manifolds

TL;DR

HH-SAE is a hierarchical model that disentangles complex manifolds to improve high-stakes domain understanding, achieving state-of-the-art results in fraud detection and knowledge synthesis.

Contribution

The paper introduces HH-SAE, a novel hierarchical autoencoder architecture that factorizes manifolds into nested tiers for enhanced interpretability and performance.

Findings

Achieves a cross-domain zero-shot AUC of 0.9156 in fraud detection.

Path ablation shows a 13.46% utility collapse without contextual subtraction.

Knowledge-steered synthesis yields a +9.9% AUPRC improvement over state-of-the-art generators.

Abstract

Rare semantic innovations in high-dimensional, mission-critical domains are often obscured by dense background contexts, a challenge we define as \textit{feature density conflict}. We introduce the \textbf{Hybrid Hierarchical SAE (HH-SAE)} to resolve this by factorizing manifolds into a nested hierarchy of \textbf{Contextual} ($L_0$), \textbf{Atomic} ($f_1$), and \textbf{Compository} ($f_2$) tiers. Evaluating across disparate manifolds, HH-SAE demonstrates superior resolution by \textbf{``fracturing'' administrative clinical labels into physiological modes} and achieving a peak \textbf{cross-domain zero-shot AUC of 0.9156 in fraud detection}. Path ablation confirms the architecture's structural necessity, revealing a 13.46\% utility collapse when contextual subtraction is removed. Finally, knowledge-steered synthesis achieves a +9.9\% AUPRC lift over state-of-the-art generators, proving that HH-SAE effectively prioritizes high-order mechanistic innovation over environmental proxies to enable high-precision discovery in high-stakes environments.