Introduction

Structured inference has become increasingly crucial for large language models. However, managing constraints can be a significant bottleneck to efficiency and scalability.

Presentation of MetaJuLS

MetaJuLS is a meta-learning solution that introduces an universal approach to constraint propagation for structured inference. This approach uses a graph attention network to form human-like parsing strategies and novel non-intuitive heuristics.

Experimental Results

MetaJuLS achieves 1.5-2.0x speedups over GPU-optimized baselines while maintaining within 0.2% accuracy of state-of-the-art parsers. The solution demonstrates rapid cross-domain adaptation with a policy trained on English parsing adapting to new languages and tasks in just 5-10 gradient steps (5-15 seconds).

Contributions

MetaJuLS contributes to Green AI by directly reducing inference carbon footprint.