IVD-SSM Uses Gated State-Space Models for Narrative Similarity
July 7, 2026
The Invariant-Variant Disentangled State-Space Model uses a Jamba-1.5-Mini backbone and a Structurally Gated Alignment head to separate superficial story elements from causal patterns. The SGA head employs a strided Macro-path to gate a high-resolution Micro-path for efficient long-context modeling.
HOW THIS AFFECTS YOU
●
researcherThis demonstrates how hybrid SSM-Transformer architectures can disentangle high-level narrative structure from low-level detail.