Multi-Task Inference Using Gradient Feedback From Subtask Labels

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing multi-task learning models fail to effectively reflect user feedback on subtask inference results across tasks, limiting accuracy improvements.

Innovation Solution

An information processing apparatus and method that acquires data and subtask labels to calculate gradients, allowing for task inference using a learned model generated by multi-task learning, thereby incorporating user feedback to enhance accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a learned model generated by multi-task learning is used for task inference, then accuracy of subtask inference can be improved, but the effect of user feedback on subtask inference results cannot be reflected in the inference of other subtasks

Engineering Contradiction:

Improveaccuracy of subtask inferenceVSAvoidability to reflect user feedback across tasks

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent implements a feedback mechanism where user feedback on subtask inference results is calculated and propagated back to the input data through gradient computation. The system calculates the gradient of the loss function with respect to the input data based on user feedback, then updates the input data using this gradient to reflect user preferences in subsequent inferences across multiple tasks.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent changes the parameters (input data) based on user feedback by computing gradients and performing optimization updates. Specifically, the system adjusts the input data representation to minimize the discrepancy between predicted and actual user preferences, thereby adapting the inference results across different tasks.

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If user feedback is incorporated into task inference using gradient calculation, then accuracy can be enhanced across multiple tasks, but computational complexity increases

Engineering Contradiction:

Improvetask inference accuracyVSAvoidcomputational complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary actions by pre-computing and storing the gradient of the loss function with respect to the input data before actual inference. This allows the system to quickly apply user feedback during inference without performing complex gradient calculations in real-time, thereby reducing computational complexity during the inference phase.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20260017541A1Information processing apparatus, method, and non-transitory computer-readable medium for task inference using machine learning with gradient based multi-task learning

Publication Date: 2026.01.15 NEC CORP

AI summary

An information processing apparatus includes at least one memory storing instructions, and at least one processor configured to execute the instructions to acquire data to be input to a first layer included in a plurality of layers forming a learned model generated by multi-task learning, acquire an inference result of a subtask, the inference result being output from a second layer, which is a layer subsequent to the first layer, by inputting the data to the first layer, acquire a subtask label corresponding to the data, calculate a gradient in the data of a function using, as inputs, the inference result of the subtask and the subtask label, and perform task inference using the data, the gradient, and the learned model. As an example, the information processing apparatus is used for decision-making assistance such as diagnosis using a medical image.