Processing machine learning techniques using a graphics processing unit

A computer-implemented process for accelerating and optimizing a machine learning technique using a graphics processing unit, comprising:

• using a general-purpose computing device to perform the following process:
• training learnable parameters of the machine learning technique to obtain trained learnable parameters;
• using the trained learnable parameters to obtain a solution to a function of the machine learning technique, obtaining a solution to the function further comprising:
• specifying a rectangular viewport contained within a given triangle in a memory of the graphics processing unit;
• using a pixel shader to compute an inner product by calculating only those pixels contained within an intersection of the rectangular viewport and the given triangle such that the pixels within the rectangular viewport become a rendering target;
• decomposing the inner product into sub-problems and performing multiple passes over the sub-problems using pixel shaders;
• using the pixel shader to perform at least one of the following:
• (a) training learnable parameters;
• (b) using the trained learnable parameters;
• and
• using the solution to the function in an interactive-use computer application.

A computer-readable storage medium having stored and encoded thereon computer-executable instructions for processing a machine learning technique using a graphics processing unit, comprising:

• loading pixel shaders onto the graphics processing unit;
• loading data onto the graphics processing unit;
• calling the pixel shaders to obtain a solution to a trained function of the machine learning technique using the graphic processing unit, obtaining the solution to the trained function further comprising:
• specifying a rectangular viewport contained within a given triangle in a memory of the graphics processing unit;
• using the pixel shaders to compute an inner product by calculating only those pixels contained within an intersection of the rectangular viewport and the given triangle such that the pixels within the rectangular viewport become a rendering target;
• decomposing the inner product into sub-problems and performing multiple passes over the sub-problems using pixel shaders, decomposing the inner product further comprising:
• creating an inner product pixel shader that will multiply a submatrix of a matrix by a subvector of a vector;
• allocating a portion of available texture maps to the matrix and a remainder of the texture maps to the vector;
• computing the inner product using the inner product pixel shader and the matrix, vector, and texture maps;
• computing an outer product using an outer product pixel shader and an indexer texture technique;
• and
• using the solution to the trained function in an interactive-use computer application.