arXiv stat.ML
· Papers
Persistent Convolution: A Topological Framework for AI Alignment Testing and Semantic Space Characterization
arXiv:2607.29008v1 Announce Type: new Abstract: Modern opaque AI models prize performance over interpretability, which makes testing difficult. However, formal statistical tests conducted on a model's embedding space can provide robust characterizations of semantic structure, concept separation, and knowledge graph ali