SBIR Phase I: Categorical Representation Learning in Artificial Intelligence

The broader impact of this Small Business Innovation Research (SBIR) Phase I project is substantial improvement in artificial intelligence. Despite advances in AI, the current state-of-the-art platforms are still weak in learning and require large amounts of training data. The proposed system creates a better AI system by exploring more relationships within complex data sets. The initial application is cancer detection in genomic data.

This Small Business Innovation Research (SBIR) Phase I project is to develop a novel relation-oriented machine learning platform. Current platforms are object-oriented, where objects (e.g., words, sequence data, etc.) are represented as feature vectors. The feature vector representation is powerful in performing tasks such as pattern recognition, classification, and regression but ineffective in learning the interrelationships of objects in great detail.

An in-depth understanding of objects is paramount in learning. Also, the meaning of objects is defined and can only be defined through their interrelationship with other objects. Based on category theory in mathematics and quantum physics, the proposed platform is relation-oriented.

It maps and preserves the interrelationship between objects in all dimensions. Based on the observed concurrences of objects and their relations, the platform fuses them to create more complex objects progressively and learns their interrelations iteratively to form a hierarchical structure associated with the dataset. This platform automatically learns the meaning of objects and the governing rules between them.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.