Learning from small amounts of data

At Evolution AI we’ve invested a lot of time and effort into overcoming the limitation of little or no training data, while training the models to state-of-the-art accuracy levels

Create more training data easily using our intuitive annotation interface

Underlying machine reading algorithms pre-trained on millions of examples

QA interface allows you to easily check model errors

API integration point for your applications, providing an easy way to connect your systems to trained models

active learning vs baseline graph

Active learning

The key idea behind active learning is that an algorithm can improve its performance with fewer training labels if it chooses the data from which it learns.

For clear illustration, the graph shows that the same algorithm trained by Evolution AI using active learning (blue) achieves the same accuracy with three times less data compared to the baseline (orange).

Transfer learning

Transfer learning is a machine learning technique where a large model trained on one set of data is applied to a new task with little labelled data.

Models that have already learned to read text in one domain consistently require less data, and end up being more accurate than those that have to learn to read from scratch.

graph comparison