This inventor holds 1 USPTO granted patent. Top assignee: Tensordyne, Inc.. Active years: 2026.
Data source: USPTO (United States Patent and Trademark Office) public patent records. Weekly synchronization. How IDiyas builds this profile
Company Filing History:
Years Active: 2026
Title: Michael Wilhelm Laraia and Logarithmic Mathematics for AI Matrix Multiplications
Introduction
Michael Wilhelm Laraia is an inventor based in Munich, Germany. His patent work addresses mathematical operations used in artificial intelligence, focusing on a matrix multiplication system that combines logarithmic-domain multiplication with linear-domain accumulation.
Latest Patents
Laraia is associated with 1 patent, titled “Method and Apparatus for Logarithmic Mathematics in AI Matrix Multiplications.” The invention describes co-designed circuits that convert linear-domain operands into a logarithmic domain for multiplication, then convert the results back into a linear domain for accumulation.
The linear-to-logarithmic conversion divides a mantissa into four non-uniform windows and uses affine mappings implemented with shift and add operations. The logarithmic-to-linear conversion divides a mantissa into four equal-width windows and applies corresponding affine mappings, also using shift and add operations. The mappings are designed as analytic inverses, with the first conversion’s window boundaries derived from the second conversion’s boundaries. This co-design aims to provide round-trip exactness apart from quantization effects and reduce systematic bias during accumulation.
Career Highlights
The available information identifies Laraia’s patent contribution in AI matrix multiplication and lists Tensordyne, Inc. as his company. The invention’s approach is designed to support efficient arithmetic through conversion circuits that rely on shift and add operations.
Collaborations
The listed coworkers associated with Laraia are Jian Hui Huang and Gary S. Goldman. The available details do not specify their individual roles in the patent or their broader professional collaborations.
Conclusion
Michael Wilhelm Laraia’s patent describes a co-designed conversion approach for logarithmic multiplication and linear accumulation, with the goal of limiting bias while supporting AI matrix computations.
Data source: USPTO (United States Patent and Trademark Office) public patent records. Weekly synchronization. How IDiyas builds this profile