NEHlo

NEHlo is a fictional open-source framework for hierarchical optimization using neural-guided search. It provides a representation for hierarchical decision problems where high-level decisions constrain lower-level choices, and uses a combination of traditional optimization backends and neural networks to guide search. It supports modeling tasks such as facility layout, job scheduling, and multi-robot planning by decomposing problems into levels and enabling communication between levels through defined interfaces.

History: Began as a research initiative in 2020 by contributors from an unnamed consortium; first public release in 2022; maintained with active contributions through 2024; community hosted on GitHub.

Technical overview: Implemented in Python with performance-critical parts in C++. Core components include a problem DSL for hierarchy, a solver orchestrator, a neural guidance module interface, and a collection of plug-in solvers (MILP, CP-SAT, heuristic search). It includes a training loop for shape of neural policies using imitation learning or reinforcement learning. It supports interoperability with standard formats like JSON and YAML.

Examples: The library illustrates a two-level problem where the upper level assigns regional production targets and the lower level schedules machinery, with NEHlo coordinating via calls between levels.

Reception and status: In academia, NEHlo is cited in optimization and AI planning literature as a framework for exploring neural-guided hierarchical optimization. It remains experimental and is not widely adopted in production settings, due to complexity and compute requirements.