Conceptually there are the following major approaches:
- Interactive (Front end)
- SDK Model Wrapper
You can use your model to pre-label data. This means predictions will show up in the Diffgram UI. This can be used to review data, improve data, and more.
There are multiple ways to send that data to Diffgram.
You can use the import wizard or the SDK or REST API
You can even update existing files multiple times.
And you can store multiple model prediction version on the same file and compare them.
Learn more Importing Introduction
JS or via Realtime API call
Integrate ML backends in an event driven way. Workflows
- Model Assisted Labeling (MAL)
- Auto Annotate
- ML Backend
- Active Learning
- AI Assist
- Data Centric AI
Active learning is when the input from the human effects the AI model. Some people feel this must be "interactive" e.g. real time updates to the model. Others see it more broadly as any type of iteration loop, even if it's over a period of days. Diffgram supports a wide variety of active learning methods, and we are continuing to add more.
Updated about 1 month ago