Handwriting recognition

A system comprising:

• one or more processors;
• and
• one or more hardware storage devices having stored thereon computer-executable instructions that, when executed by the one or more processors, configure the system to perform at least the following:
• receive an input image comprising handwriting;
• generate an embedding based on the input image using first network comprising a convolutional neural network, the convolutional neural network comprising a plurality of convolution layers and one or more pooling layers;
• the convolutional neural network comprising a rectified linear unit after one or more of the plurality of convolution layers;
• generate one or more character predictions based on the embedding using a second network comprising a cascade of concatenated convolution layers, the cascade of concatenated convolution layers comprising one or more subnetworks configured to perform at least one of:
• receive the embedding as an input,
• downsample the embedding to a feature map,
• flatten the feature map to a vector, and
• generate a character prediction from the vector using a dense neural network.