One of the most complex steps in creating an artificial intelligence model is to find enough example data, from which the neural network can learn, and manually annotate it.
Logiroad.ai has put in place several strategies to address this issue.
- Learning transfer allows to reuse neural networks that already have a good generic understanding of shapes, objects, or language, if possible already in a close domain, and then to specialize them on the task to which they will be dedicated.
- Data augmentation, which allows from a data set, to generate other similar examples, but a little different, to allow the network to learn more.
- Active learning, which makes it possible to select a subset of data that will be the most likely to add value to the learning, which can reduce the need for data annotations by a factor of 10.