INFORMATION PROCESSING APPARATUS, METHOD, AND NON-TRANSITORY COMPUTER-READABLE MEDIUM FOR TASK INFERENCE USING MACHINE LEARNING WITH GRADIENT BASED MULTI-TASK LEARNING

Smart overview of the Invention

An innovative information processing apparatus is designed to enhance task inference accuracy using a multi-task learning model. This apparatus comprises a memory storing instructions and a processor executing these instructions. It processes data through a series of layers within a learned model, utilizing gradient-based multi-task learning. The system is particularly beneficial for decision-making tasks, such as medical image diagnosis.

Technical Field

The apparatus operates within the domain of information processing, employing advanced machine learning techniques. Multi-task learning is leveraged to improve the accuracy of models by simultaneously addressing multiple related tasks. This approach aims to refine task inference by incorporating user-modified inference results of auxiliary subtasks, which traditional methods like those in JP 2021-174428 A fail to accommodate effectively.

Methodology

The apparatus functions by acquiring input data for the initial layer of a multi-layered learned model. It then retrieves inference results from a subsequent layer, identifies corresponding subtask labels, and calculates gradients using these inputs. This process allows for refined task inference by integrating the data, gradient, and learned model, thus enhancing the inference accuracy.

Implementation

The implementation involves a structured sequence of operations: data acquisition, inference result retrieval, subtask label acquisition, gradient calculation, and task inference execution. These steps are facilitated by a non-transitory computer-readable medium storing a program that directs a computer to perform these processes, ensuring efficient and accurate task inference.

Benefits

The described information processing apparatus provides significant improvements in task inference accuracy by utilizing a learned model from multi-task learning. This approach allows for more precise decision-making support, particularly in fields requiring high accuracy, such as medical diagnostics. The ability to incorporate user-modified inference results into the learning process distinguishes this method from previous techniques, offering a robust solution for complex inference tasks.