Anton Vsevolodovich Trusov

Title: Anton Vsevolodovich Trusov: Inventor in Quantized Neural Networks

Introduction

Anton Vsevolodovich Trusov is an inventor based in Moscow, Russia, whose patent work addresses efficient computation in quantized neural networks. He is associated with Smart Engines Service LLC and is credited with 1 patent.

Latest Patents

Trusov’s patent, “Almost-Indirect Convolution in Quantized Neural Networks,” describes a method for carrying out convolution operations in quantized neural networks. The method allocates and initializes an indirection buffer with pointers to blocks of values in an input tensor representing an image. Across iterations, blocks of input rows are packed into a cache using the buffer, and a convolution kernel processes each packed block to produce outputs. Those outputs are accumulated into an output tensor representing the resulting image.

Career Highlights

The patent reflects Trusov’s work on optimizing neural-network computation through data organization and iterative matrix multiplication. His disclosed approach uses an indirection buffer and cache packing to support convolution in quantized neural network layers.

Collaborations

Trusov is associated with Smart Engines Service LLC. His listed coworkers include Elena Evgenyevna Limonova and Sergey Aleksandrovich Usilin.

Conclusion

Anton Vsevolodovich Trusov’s patent focuses on a practical method for performing convolution in quantized neural networks, highlighting his contribution to efficient machine-learning computation.