was successfully added to your cart.

Natural Language ProcessingAnnotation Lab

Model Tuning and Transfer Learning in the Annotation Lab

By October 4, 2021October 6th, 2021No Comments

We’re very excited to announce that Annotation Lab v.2.1.0 has been enriched with┬ámore advanced model training options and the great Transfer Learning.

Training

While training a model within the Annotation Lab, Project Owner or Manager can specify the Annotator Approach they want to use. Two options are currently available:

  • NerDL offered by Spark NLP library – e.g Open Source
  • MedicalNER offered by Spark NLP for Healthcare library -e.g. Licensed.
  • Train Open Source Models

 

Screen Shot 2021-09-23 at 19 17 48

 

The auto-refresh feature was added for Live training logs. Those can be accessed via the small icon next to the training button.

 

 

Transfer Learning

While using the Transfer Learning feature, it is often the case that the labels present in the current Project configuration (thus also in the annotations defined so far in the project) might be slightly or completely different from what of the base model (which is used for Transfer Learning). This can lead to unexpected Transfer Learning results, so in such a case, a proper warning about what is different is shown when selecting the model in the “Advanced Training Options” box.

 

 

Models Hub

This version reduces the internet dependency of the product on some default models and embeddings. The Annotation Lab image used for deployments now includes several basic Spark NLP components (models and embeddings) so that system admins are expected to face fewer issues during the installation in air-gapped or enterprise environments.

Note: Access to the Models Hub page and downloading Spark NLP and Spark NLP for Healthcare models and embeddings requires an internet connection.