Intelligence Augmentation - overview
The Intelligence Augmentation team aims to reduce the time to develop and deploy machine learning solutions by accelerating the creation of semantic assets and maximizing the outcome of intelligent interactions with subject matter experts. Our technology enables cognitive collaboration, rapid co-learning - in any domain and any language - between multiple stakeholders and subject matter experts where the computer is an equal participant.
Both a challenge and key to success for machine learning systems is the availability of reliable annotated data or formalized knowledge (e.g. in the form of an ontology) to rapidly create and then sustain the data annotation process. The dependency on fresh and reliable annotated data can be a huge stumbling block in creating and maintaining machine learning systems, particularly as knowledge evolves over time.
The research explores efficient knowledge construction (i.e., dictionary and ontology) with human-in-the-loop technology that presumes the availability of a subject matter expert while optimizing their time for rapid development. Key innovations include:
- Augmenting concept expansion using statistical methods with human feedback to tune extraction patterns
- Fast start using Linked Data to inform the extraction of concepts from unstructured text
- Language and grammar independence
- Adaptivity to language drift