Convolutional neural network architecture with excellent performance on small data for image recognition

Computerised Image recognition offers a fantastic opportunity to improve a firm’s efficiency as search, classification and analytic tasks are sped up and require less human labour. However, in situations where data, time or computing power are scarce, the effectiveness of image categorisation through machine learning may hamper. Prof. Dr. Ir. A. W. M. Smeulders and his research team invented a solution for these situations.  Their CNN architecture outperforms state-of-the-art convolutional neural networks on classification accuracy and computation time when based on small datasets, as tests on CIFAR-10 and MNIST data show. Their receptive field network architecture achieves: higher accuracy on small datasets; faster learning on small datasets; and equal performance on large datasets.

Click for more details on this collaboration opportunity