These three metrics are generally used
Continuous studying is a subfield of Machine Studying that offers with incrementally coaching neural networks on frequently arriving knowledge. Crucially, the info can’t be saved solely and sometimes instances no samples in any respect may be carried over from previous duties. As a result of the networks are solely optimized on the presently out there knowledge, they overwrite the previous parameters. In overwriting them, previous information often is destroyed, i.e. forgotten.
To benchmark continuous studying, and catastrophic forgetting, a number of analysis metrics are utilized in continuous studying analysis. On this article, I’ll element the three mostly used metrics. Whereas I’ll be utilizing classification for instance, the metrics equally apply to different issues, e.g. regression. In case you might be new to the subject of continuous studying, I like to recommend you learn my earlier two articles to get a deeper understanding of the subject. As I’ve accomplished earlier than, I’ll be offering studying suggestions to discover the subject additional on the finish of the article.
Common Accuracy
The primary generally used metric is common accuracy, typically abbreviated as ACC. Because the identify signifies, it measures the (test-set) accuracy of every job, after which computes the common over the duty particular accuracies. Formally is…
