Back to results
US 11,902,369 B2Granted
Autoencoder, data processing system, data processing method and non-transitory computer readable medium
AssigneePreferred Networks, Inc.
InventorsKen Nakanishi, Shinichi Maeda
Life of this patent
Estimating the term…
Abstract
An autoencoder includes memory configured to store data including an encode network and a decode network, and processing circuitry coupled to the memory. The processing circuitry is configured to cause the encode network to convert inputted data to a plurality of values and output the plurality of values, batch-normalize values indicated by at least two or more layers of the encode network, out of the output plurality of values, the batch-normalized values having a predetermined average value and a predetermined variance value, quantize each of the batch-normalized values, and cause the decode network to decode each of the quantized values.
Claims27
A method for compressing data comprising:
- acquiring, by processing circuitry, data to be compressed;
- and
- generating, by the processing circuitry, first compressed data based on an output obtained by inputting the data into a first neural network model;
- wherein a size of the first compressed data is smaller than a size of the data, and the first compressed data includes at least a feature regarding a first resolution and a feature regarding a second resolution, the second resolution is higher than the first resolution,
- wherein, in a device for decompressing data, the feature regarding the first resolution is inputted into one layer of a second neural network model and the feature regarding the second resolution is inputted into another layer of the second neural network model, the another layer being different from the one layer,
- wherein the first neural network model has at least one layer and another layer which is deeper than the one layer of the first neural network model, the feature regarding the first resolution is generated based on an output of the another layer of the first neural network model, and the feature regarding the second resolution is generated based on at least a first portion of an output of the one layer of the first neural network model, and
- wherein at least a second portion of the output of the one layer of the first neural network model is inputted into the another layer of the first neural network model.
4 more elements
A method for decompressing data comprising:
- acquiring, by processing circuitry, compressed data;
- and
- generating, by the processing circuitry, decompressed data by inputting the compressed data into a neural network model which has at least one layer and another layer;
- wherein a size of the decompressed data is larger than a size of the compressed data, and the compressed data, including at least a feature regarding a first resolution and a feature regarding a second resolution which is higher than the first resolution, is generated based on an output of another neural network model that has inputted data to be compressed,
- wherein the feature regarding the first resolution is inputted into the one layer and the feature regarding the second resolution is inputted into the another layer,
- wherein the another neural network model has at least one layer and another layer which is deeper than the one layer of the another neural network model, the feature regarding the first resolution is generated based on an output of the another layer of the another neural network model, and the feature regarding the second resolution is generated based on at least a first portion of an output of the one layer of the another neural network model, and
- wherein at least a second portion of the output of the one layer of the another neural network model is inputted into the another layer of the another neural network model.
4 more elements
A method for learning a model comprising:
- inputting, by processing circuitry, first data into a first neural network model to generate second data which is compressed data of the first data;
- inputting, by the processing circuitry, the second data into a second neural network model to generate third data which is decompressed data of the second data, the second data including a plurality of features each regarding different resolutions;
- and
- learning, by the processing circuitry, the first neural network model and the second neural network model based on difference between the first data and the third data,
- wherein a size of the second data is smaller than a size of the first data, and the second data includes at least a feature regarding a first resolution and a feature regarding a second resolution, the second resolution is higher than the first resolution,
- wherein the feature regarding the first resolution is inputted into one layer of the second neural network model and the feature regarding the second resolution is inputted into another layer of the second neural network model, the another layer being different from the one layer,
- wherein the first neural network model has at least one layer and another layer which is deeper than the one layer of the first neural network model, the feature regarding the first resolution is generated based on an output of the another layer of the first neural network model, and the feature regarding the second resolution is generated based on at least a first portion of an output of the one layer of the first neural network model, and
- wherein at least a second portion of the output of the one layer of the first neural network model is inputted into the another layer of the first neural network model.
5 more elements
A device for compressing data comprising:
- at least one memory;
- and
- at least one processor configured to:
- acquire data to be compressed;
- and
- generate compressed data based on an output obtained by inputting the data into a first neural network model;
- wherein a size of the compressed data is smaller than a size of the data, and the compressed data includes at least a feature regarding a first resolution and a feature regarding a second resolution, the second resolution is higher than the first resolution,
- wherein, in a device for decompressing data, the feature regarding the first resolution is inputted into one layer of a second neural network model and the feature regarding the second resolution is inputted into another layer of the second neural network model, the another layer being different from the one layer,
- wherein the first neural network model has at least one layer and another layer which is deeper than the one layer of the first neural network model, the feature regarding the first resolution is generated based on an output of the another layer of the first neural network model, and the feature regarding the second resolution is generated based on at least a first portion of an output of the one layer of the first neural network model, and
- wherein at least a second portion of the output of the one layer of the first neural network model is inputted into the another layer of the first neural network model.
7 more elements
A device for decompressing data comprising:
- at least one memory;
- and
- at least one processor configured to:
- acquire compressed data;
- and
- generate decompressed data by inputting the compressed data into a neural network model which has at least one layer and another layer;
- wherein a size of the decompressed data is larger than a size of the compressed data, and the compressed data, including at least a feature regarding a first resolution and a feature regarding a second resolution which is higher than the first resolution, is generated based on an output of another neural network model that has inputted data to be compressed,
- wherein the feature regarding the first resolution is inputted into the one layer and the feature regarding the second resolution is inputted into the another layer,
- wherein the another neural network model has at least one layer and another layer which is deeper than the one layer of the another neural network model, the feature regarding the first resolution is generated based on an output of the another layer of the another neural network model, and the feature regarding the second resolution is generated based on at least a first portion of an output of the one layer of the another neural network model, and
- wherein at least a second portion of the output of the one layer of the another neural network model is inputted into the another layer of the another neural network model.
7 more elements
Dependencies are inferred from the claim text.