Inference apparatus, learning apparatus, inference method, and learning method

An inference apparatus comprising:

• a memory;
• and
• a hardware processor coupled to the memory and configured to:
• acquire at least one control parameter of second machine learning model, the second machine learning model having a size smaller than a size of a first machine learning model input to the inference apparatus;
• decompose, by using a tensor decomposition method, a tensor of weight coefficients of the first machine learning model into two or more decomposed tensors;
• set a width of each of the decomposed tensors and set sizes of the second machine learning model in accordance with the at least one control parameter;
• change the first machine learning model to the second machine learning model represented by the two or more decomposed tensors each having a set size, and
• perform inference in response to input data by using the second machine learning model, wherein
• the tensor of the weight coefficients is a matrix of the weight coefficients,
• the decomposed tensors are two or more matrices each constituted by R basis vectors, wherein R is an integer value of a number of basis vectors,
• the hardware processor sets the width of each decomposed tensor by selecting r basis vectors (r≤R) from the R basis vectors, wherein r is an integer value of a number of basis vectors, and
• the hardware processor changes the first machine learning model to the second machine learning model represented by the two or more decomposed tensors each having the set width.