Widest-Path Reachability Fields for Connectivity-Preserving Slender Structure Segmentation
A differentiable Max-Min reachability objective that redirects training gradients toward connectivity bottlenecks in slender-structure segmentation.
Open-Ended Scenario Reasoning for Specialist Model Adaptation
A risk-gated Bayesian adaptation framework that uses open-ended scenario evidence to correct frozen specialist models without retraining.
LLM-Guided Task-Semantic Field Factorization for Industrial Process Forecasting
An offline LLM-guided input factorization that lets each numerical window activate task-aware variable semantics before a standard forecasting backbone.
LLM-Guided Measurement Credibility Correction for Trustworthy Industrial Process Inference
A pre-inference correction framework that uses measurement semantics to build independent references for locally inconsistent industrial signals.