Brain Builder Knowledge Base
Training your Brain
Direct link to topic in this publication:
Brain Builder uses Neurala's patented Lifelong-AITM technology to build Recognition Brains one image at a time.
This process has numerous advantages over traditional DNN training processes, which require large amounts of labeled images and cannot be easily updated with additional data after the initial training is complete.
Once you have created a project and uploaded data, it's time to get started training your Brain!
Tagging = Training
Brain Builder learns as you tag, so the tagging and training process is one and the same.
Navigate to your Recognition Dataset and click Resume Tagging.
This will take you to the Workspace, starting with the first image that has not yet been tagged.
From the list of available tags to the right of the image, select the correct tag for the image. Or, if you don’t see the class you need, click the + button next to Classes and add it to the list.
Click the Next Image arrow, and then select Save & Continue. Brain Builder will save your tag and the Brain will learn this image and its associated tag.
When you're tagging images for your Recognition Brain, there are a few details you should keep in mind:
Try to balance the number of images for each of the classes you are training. They don't have to match exactly, but it's better for each class to have a similar number of images than for one to have far more or fewer than the others.
For the best results, images should clearly correspond with a single class. Brain Builder will not deliver accurate results if it's trained on images that are associated with multiple classes. (For example, if you're training on "Dog" and "Person" you shouldn't use images that include both people and dogs.)
For more tips and best practices, be sure to review Best Practices: Recognition Brain Training.
Keep tagging images in this way until your Brain starts making predictions as to which tag the image should have.
When you first start to tag images and train your Brain, you'll see a message that the Brain isn't able to make a prediction yet.
After you have tagged enough images, you'll see Brain Builder's Top Predictions for the image appear above the Class list.
Each prediction is displayed with a score showing the strength of the prediction.
Brain Builder will likely make some wrong predictions or will be unable to make a prediction for some images. This is normal. Brain Builder needs to learn enough images before it can make good predictions.
When Brain Builder makes a correct prediction, clicking that prediction and saving it will help reinforce Brain Builder's knowledge of that class. If Brain Builder makes an incorrect prediction, you should correct it by tagging the image with the correct class and saving that knowledge into the Brain.
As the Brain learns, the tagging process becomes as easy as selecting the top prediction and saving it. Within minutes, you will have trained your Brain.
Is That It?
It's really that easy. Upload images, tag them, and the Brain learns as you go. Unlike most other vision AI solutions, a Brain Builder Recognition Brain can be trained in under an hour.
Next: Retraining your Brain >>