Dynamic Spatio-Temporal Attention Bias Generator
Dynamic Spatio-Temporal Attention Bias Generator
Auto-generated stub. Edit this file to add more details.
A mechanism that injects persistent functional graphs and transient nodal states into LLM attention layers to steer sequence modeling for spatio-temporal data.
Why It Matters
This provides a generalized method for injecting relational graph priors into transformer-based attention, which is critical for applying LLMs to non-linguistic structured domains.
Evidence
Our core innovation is a Dynamic Spatio-Temporal Attention Bias Generator that synthesizes a persistent functional graph with transient nodal states to explicitly steer the LLM’s attention.
Related Papers
Metadata & Links
- created_at
- 2026-05-15T06:15:23Z