Deep learning models rely on large annotated datasets in order to obtain good results. However, in many domains such as the security and healthcare domains there is a lack of large annotated datasets due to the high cost and long time needed to acquire and/or label relevant data. Transfer learning allows features that are learned from a large annotated dataset in a source domain to be transferred to a target domain only a small labelled dataset is available.
WHAT DOES TNO OFFER ON TRANSFER LEARNING?
- TNO offers transfer learning methods for visual object recognition that make it possible to exploit already available but less representative data.
- TNO offers methods that improve on traditional transfer learning methods by not only transferring features but also labels from the source domain.
TNO SPIE 2016 Paper: “Object recognition using deep convolutional neural networks with complete transfer and partial frozen layers” (https://doi.org/10.1117/12.2241177)