The Solution of The Sudoku Games: Exploration of The Potential of Cnn in Logical Reasoning Problems

Logical reasoning problems, such as Sudoku, have long been considered a challenge for traditional artificial intelligence techniques. While these problems require a high level of pattern recognition and deductive reasoning, the application of deep learning, particularly Convolutional Neural Networks (CNNs), in solving such tasks has not been extensively explored. Recent advancements in neural networks have shown promise in handling complex reasoning tasks, yet their potential in logical problem-solving domains remains largely untapped. However, exploring the potential of CNN in logical reasoning problems, such as Sudoku, remains an under-explored area. The research uses a dataset of one million Sudoku games from Kaggle. Data is preprocessed and fed into a CNN model. The model is evaluated using loss and accuracy metrics, trained with a batch size of 64 for up to 100 epochs, and optimized with Adam optimizer. The model achieves a test accuracy of 0.9632, indicating some generalization ability. This study is significant as one of the early efforts to apply CNN to logical reasoning. It provides a basis for future research, highlighting both the potential of CNN in this area and the challenges that need to be addressed for further improvement.