发表机构
Institute for Machine Learning and Analytics (IMLA), Offenburg University; University of Mannheim（奥芬堡大学机器学习和分析研究所; 曼海姆大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对空间推理难题，提出自适应神经细胞自动机，利用可变形卷积动态调整感知野并迭代推理二维空间关系，在数独和迷宫等图像谜题上取得最先进结果。
AI 中文摘要
许多现代学习方法在空间推理任务上仍面临挑战，即它们缺乏利用感知实体的几何信息及其相互空间关系来解决问题的能力。我们提出了一种新颖的自适应神经细胞自动机（aNCA）架构，该架构使用可变形卷积动态调整感知视野，并在网格状数据结构（如图像）上迭代推理二维空间关系。在公开基准上的实验结果表明，该方法在解决基于图像的谜题（如数独或迷宫中最短路径查找）时，取得了最先进的可理解结果，并具有较高的泛化能力。
英文摘要
Many modern learning approaches are still struggling with spatial reasoning tasks, i.e. they lack the ability to utilize geometric information of perceived entities and their spatial relation to each other to solve problems. We introduce a novel Adaptive Neural Cellular Automata (aNCA) architecture which uses deformable convolutions to dynamically adapt the perceptive field and iteratively reason over 2D spatial relations on grid-like data structures (e.g. images). Empirical results on public benchmarks show state of the art comprehensible results with high generalization abilities for solving image based puzzles like Sudoku or finding the shortest path in a maze.
