JPWO2021095361A5 - - Google Patents

Claims (8)

A storage unit (12) that stores a trained neural network model, in which the weighting coefficient is defined by a probability distribution and includes a probability layer that propagates the mean and variance of the probability distribution of the output value to the subsequent stage.
Input unit (10) for inputting data and
An inference unit (11) that analyzes the data input in the input unit by inference by the neural network model, and
An output unit (13) that outputs the analysis result by the inference unit, and
An arithmetic unit (1). In the layer where the mean and variance of the probability distribution are input, the first mode to calculate the variance to be output assuming that the values input from the previous layer are independent, and the values input from the previous layer are non-independent. The arithmetic unit according to claim 1, further comprising a second mode for calculating the variance to be output on the assumption that A storage unit (12) that stores the trained neural network model,
Input unit (10) for inputting data and
An inference unit (11) that analyzes the data input in the input unit by inference by the neural network model, and
An output unit (13) that outputs the analysis result by the inference unit, and
Equipped with
The neural network model has one probability layer composed of a combination of a given probability distribution and a weighting coefficient set by learning, and the other layers define the weighting coefficient with a definite value.
The inference unit analyzes the input data based on the value appearing in the probability layer and the probability distribution of the weighting coefficient when the data input by the input unit is applied to the neural network model. (1). The arithmetic unit according to claim 3, wherein the probability layer is arranged at a position closest to the output layer. The input unit inputs image data, voice data, or text data, and inputs the image data, the voice data, or the text data.
The arithmetic unit according to any one of claims 1 to 4 , wherein the inference unit classifies the image data, voice data, or text data into a plurality of classes and obtains reliability classified into the classes. The input unit inputs an image taken by the camera (20) of the autonomous driving vehicle, and inputs the image.
The reasoning unit classifies the objects shown in the image into classes and infers their reliability.
The arithmetic unit (2) according to any one of claims 1 to 4 , wherein the output unit outputs the class of the object and the reliability data to the vehicle control unit (21). A trained model of a neural network that makes a computer function to analyze input data.
The weighting coefficient is defined by the probability distribution and consists of a neural network model that includes a probability layer that propagates the mean and variance of the probability distribution of the output values later.
A trained model that analyzes the input data based on the mean and variance of the probability distribution of the output values propagated to the output layer when the input data is applied to the neural network model. A trained model of a neural network that makes a computer function to analyze input data.
It has one probability layer composed of a combination of a given probability distribution and a weighting coefficient set by learning, and the other layer is composed of a neural network model in which the weighting coefficient is defined by a definite value.
A trained model that analyzes the input data based on the value appearing in the probability layer and the probability distribution of the weighting coefficient when the input data is applied to the neural network model.