Method for determining explainability mask by neural network, system and medium

A non-transitory computer-readable storage medium on which is stored data representing a computer program including instructions for determining an explainability mask for a classification of an input image by a trained neural network, the trained neural network being configured to determine the classification of the input image by determining a latent representation of the input image at an internal layer of the trained neural network and by determining the classification of the input image from the latent representation of the input image, determining the classification comprising determining a classification score of the input image, the computer program, when executed by a processor system, causing the processor system to perform the following steps:

• accessing the trained neural network;
• obtaining the input image and the latent representation of the input image;
• initializing a mask for indicating modifications to the latent representation;
• updating the mask by iteratively adjusting values of the mask to optimize an objective function, the objective function including i) a modification component indicating a degree of the modifications indicated by the mask and ii) a classification score component, the classification score component being determined by:
• applying the modifications indicated by the mask to the latent representation to obtain a perturbed latent representation, and
• determining the classification score of the perturbed latent representation according to the trained neural network;
• scaling the mask to a spatial resolution of the input image to obtain the explainability mask;
• and
• outputting the explainability mask.

A system for determining an explainability mask for a classification of an input image by a trained neural network, the trained neural network being configured to determine the classification of the input image by determining a latent representation of the input image at an internal layer of the trained neural network and by determining the classification of the input image from the latent representation of the input image, the determining of the classification including determining a classification score of the input image, the system comprising:

• a data interface configured to access the neural network and obtain the input image and the latent representation of the input image;
• a processor subsystem configured to:
• obtain the input image and the latent representation of the input image;
• initialize a mask for indicating modifications to the latent representation;
• update the mask by iteratively adjusting values of the mask to optimize an objective function, the objective function including:
• i) a modification component indicating a degree of modifications indicated by the mask, and ii) a classification score component, the classification score component being determined by:
• applying the modifications indicated by the mask to the latent representation to obtain a perturbed latent representation;
• and
• determining the classification score of the perturbed latent representation according to the trained neural network;
• scale the mask to a spatial resolution of the input image to obtain the explainability mask;
• and
• output the explainability mask.