The fundamental difference of our approach is that we focus on the data representation, not on the algorithm like the mainstream machine learning approaches do. Our Semantic Folding theory describes a new data representation called Semantic Fingerprint. It corresponds to the biological way to represent language information in the human brain: a sparse, distributed vector which encodes all meanings and contexts of a given text — a word, a sentence, a paragraph or even a 200-page book, choosing from a bundle of 16,000 semantic features. Semantic fingerprints are computed using set-theory and geometry instead of statistics and probability theory. Our approach combines both high computing efficiency and high precision — a paradigm change in an era where state-of-the-art models always impose a trade-off.
