Continuous active learning is an iterative form of technology-assisted review. Rather than training on a fixed seed set and then classifying the remainder, the model updates after each batch of reviewer decisions and reranks the unreviewed population accordingly.
Review proceeds in descending order of predicted responsiveness, and the process typically stops when newly surfaced documents are consistently non-responsive, validated by statistical sampling.
Alternative Names:
CAL, Active Learning Review
Why it Matters?
CAL has largely displaced earlier seed-set approaches because it requires no separate training phase and adapts as understanding of the issues evolves. It also front-loads the valuable documents, so case teams see the important material within days rather than after the full review concludes, which changes how quickly a matter can be assessed.
Frequently Confused with
Related terms
Frequently asked questions
How does CAL differ from earlier predictive coding?
When does a CAL review stop?
