Information processing device, information processing program, and information processing method

By using a convolution kernel with alternating components to fit feature distributions, the method enhances machine learning model accuracy while minimizing the need for additional training data, addressing the cost and labor challenges of data annotation.

WO2026133848A1PCT designated stage Publication Date: 2026-06-25OMRON CORP
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
OMRON CORP
Filing Date
2025-11-20
Publication Date
2026-06-25

Smart Images

  • Figure JP2025040590_25062026_PF_FP_ABST
    Figure JP2025040590_25062026_PF_FP_ABST
Patent Text Reader

Abstract

The present invention improves the accuracy of a machine learning model while suppressing training data creation cost. A second model (Mds) subjects input data (vi) to a convolution operation to output second data (vadp). On the basis of first data (vr) from a first model (Mdf) and the second data (vadp), a processing unit (11) calculates output data (vo). A kernel (K) of the convolution operation is constituted using at least one kernel. A non-zero component and at least one zero component appear alternately in each direction of at least one dimension of three or more dimensions (d1-d3) in each of the at least one kernel.
Need to check novelty before this filing date? Find Prior Art