Patent · US Active

Autonomic vertical deepening in neural networks transfer learning

US11663480B2 · kind B2 · utility

1Cited by

0References

19Claims

0Family size

Assignee

International Business Machines Corporation · US

Inventors

Aaron K. Baughman · Silver Spring, US
Michael Behrendt · Randersacker, DE
Shikhar Kwatra · San Jose, US
Craig M. Trim · Ventura, US

Key dates

Filing date	Nov 15, 2019
Grant date	May 30, 2023
Priority date	—
Expiry date	Feb 13, 2041

Classification

Technology area (CPC G)Physics
CPC primaryG06N5/022
WIPO fieldComputer technology
WIPO sectorElectrical engineering

Abstract

An autonomic function executing in an artificial intelligence environment determines that a fused model responsive to a new problem space has below a threshold level of accuracy in the new problem space. A spliced layer in the fused model is autonomically cloned, the spliced layer having been extracted from a second model and inserted at a location in the fused model. The cloned layer is autonomically inserted at a second location in the fused model. An automatically constructed vector transformation transforms an output vector of a previous layer in an immediately previous location in the model relative to the second location. The cloned layer is automatically fused in the fused model using the transformed output vector as input to the cloned layer, forming a deep fused model that has a revised accuracy that is higher than the accuracy relative to an ontology of the new problem space.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.