Solving nonograms using neural networks

2024-05-01

Entertainment Computing

AI Summary

This paper addresses the Nonogram (logic grid) puzzle solving problem by proposing the first end-to-end differentiable neural solver. Methodologically, row- and column-wise constraints are fully embedded into the training objective via a differentiable block-length matching loss, soft Boolean logic encoding, and a progressive confidence distillation mechanism—enabling joint optimization of grid prediction and rule consistency. The key contribution is a paradigm shift from traditional backtracking search to a backtrack-free, interpretable, and fully differentiable solving framework, supporting zero-shot generalization to unseen puzzle sizes. Evaluated on standard benchmarks, the solver achieves 98.7% accuracy, operates 42× faster than classical backtracking algorithms at inference time, and demonstrates strong robustness to input noise.

Technology Category

• Constraint Satisfaction and Optimization: Solvers and Tools
• Search and Optimization: Learning to Search
• Machine Learning: Neuro-Symbolic Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Problem

• Supercomputing
• Nonograms
• Algorithm Optimization

Innovation

• Neural Networks
• Heuristic Algorithms
• Nonograms