Neural network recogntion and training method and apparatus

A processor implemented recognition method comprising:

• inputting input data respectively to each of a first neural network portion of a neural network trained using a first loss function and a second neural network portion of the neural network trained using a second loss function;
• respectively normalizing each of an output of the first neural network portion and an output of the second neural network portion based on a reference level;
• obtaining a weighted average of the normalized output of the first neural network portion and the normalized output of the second neural network portion;
• and
• indicating a recognition result of the neural network based on the obtained weighted average.

A neural network apparatus comprising:

• one or more processors configured to:
• input input data to a neural network;
• determine respective mapping functions corresponding to a multiclass output of the neural network in association with the input data, including determining a mapping function of a first class and a mapping function of a second class;
• acquire a result of a loss function including a first probability component that changes correspondingly to a function value of the mapping function of the first class and a second probability component that changes contrastively to a function value of the mapping function of the second class;
• determine a gradient of loss corresponding to the input data based on the result of the loss function;
• and
• update a parameter of the neural network based on the determined gradient of loss for generating a trained neural network based on the updated parameter,
• wherein the function value of the mapping function of the first class is determined by inputting the input data to the mapping function of the first class, and
• wherein the function value of the mapping function of the second class is determined by inputting the input data to the mapping function of the second class.

An apparatus comprising:

• one or more processors configured to execute instructions;
• and
• a memory storing the instructions that, when executed by the one or more processors, configure the one or more processors to:
• generate a trained neural network by iteratively training an in-training neural network using a contrastive loss based on:
• a first probability component that changes correspondingly to a first classification result of the in-training neural network for a first class, of multiple classes classifiable by the in-training neural network, with respect to training data and the first class;
• and
• a second probability component that changes contrastively to a second classification result of the in-training neural network for a second class, of multiple classes classifiable by the in-training neural network, and the second class.

An apparatus for use with a contrastive loss trained neural network or a different contrastive loss trained neural network, respectively having characteristics of having been trained using a loss based on a first probability component that changes correspondingly to a classification result for a first class of multiple class results of a corresponding neural network, and a second probability component that changes contrastively to a classification result for a second class of the multiple class results, the apparatus comprising:

• a memory storing, as respective neural networks, at least two of the contrastive loss trained neural network, the different contrastive loss trained neural network, or a cross entropy loss trained neural network;
• and
• one or more processors configured to:
• perform an inference operation by implementing the respective neural networks with respect to input information;
• and
• normalize, based on a reference level, respective results of the implemented respective neural networks and generate a result of the inference operation based on the normalized respective results.