Overview

In order to enable any AI-based solution, organizations need to perform comprehensive machine-training to build domain-specific AI models required for solving complex automation projects within an enterprise. This training is an expensive and time-consuming process comprised of selecting specific model types, preparation of training data and transfer learning.

TRAIN automatically creates the most appropriate AI models by leveraging a combination of unsupervised and subject matter expert-infused learning, significantly reducing learning cycle time from several months to just a few hours.

Solution Overview

Integrated with Azure, TRAIN enables data scientist and ML Engineers to seamlessly mine knowledge from large volumes of unstructured text-based content to build AI models without the need for pre-compiled ontologies or labelled training data, in a fraction of the time and expense.

TRAIN requires no manual-labelling or preprocessing of training data, rather, institutional contents are consumed as-is, with minimal manual curation.

Corpus Building

TRAIN builds and expands its own corpus-based on user-defined themes and their descriptions by automatically pooling for representative data from enterprise repositories or from public sources.

Training Data

TRAIN automatically labels large volumes of textual contents using semantic and language modelling techniques. Unsupervised long-text annotation, based on deep learning, tags textual content at the document, paragraph and sentence levels.

Knowledge Graph and Model Development

Using the Auto-Ontology feature, the users can now digitize their documents into a knowledge graph without the need of manually scripted ontologies. Along with the ontologies, TRAIN also produces a set of context models and language models used for building multi-label text classification schemes. The platform continuously learns from user feedback to improve the accuracy of its output. Easy to use training pipeline provides for validation by internal subject matter experts, allowing for knowledge blending between machine and expert.

TRAIN, an automated machine teaching platform

by Parabole