Incremental learning versus rebuilding the model

Some Machine Learning algorithms, such as those for neural networks, typically require rebuilding the entire model when you add or modify any training sample. Other algorithms support incremental training, which means you can add sample information directly to the model without having to rebuild the model from all samples. Therefore, incremental training and learning is a better choice for document classification or extraction.

TotalAgility supports incremental learning, which directly reflects upon its usability and performance. In Transformation Designer, an Administrator can add a new document to the extraction model in less than a second, while rebuilding the entire model would take minutes or hours. Incremental learning also accelerates Online Learning and often the knowledge learned from operator changes is available to improve accuracy as soon as the new job is created, rather than hours later.