Multitask Neural Network Model Sharing Hidden Layers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Task-specific machine learning models are inefficient for performing multiple tasks, leading to high latency, memory footprint, and power consumption in applications like smart cameras, and require extensive curated training data for each task.
Innovation Solution
A machine-learned multitask search model is developed, which includes candidate nodes that can perform various tasks, with task controller models generating routings for task submodels to optimize performance across multiple tasks, reducing the need for individual task-specific models and training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If task-specific machine learning models are used for each task, then task performance accuracy is improved, but device complexity and resource consumption increase
Solution Approach 1:
The patent merges multiple task-specific models into a single unified model that can perform multiple tasks. The model includes multiple task-specific output layers sharing common hidden layers, allowing a single model to handle classification, detection, and segmentation tasks simultaneously, thereby reducing overall system complexity while maintaining task performance.
Solution Approach 2:
The unified model is designed with universal functionality to perform multiple machine learning tasks through a single architecture. By sharing common feature extraction layers across different task outputs, the model achieves multi-functionality without requiring separate specialized models for each task type.
2Reliability
If task-specific machine learning models are deployed, then task accuracy is improved, but memory footprint increases
Solution Approach 1:
The patent combines multiple task-specific models into one unified model where common features are extracted once and shared across all tasks. This sharing of hidden layers and parameters significantly reduces the total memory footprint compared to storing separate models for each task, while maintaining the accuracy needed for classification, detection, and segmentation.
3Adaptability or versatility
If multiple task-specific models are used, then comprehensive task coverage is improved, but power consumption increases
Solution Approach 1:
The unified model processes input data through shared computational layers once, then branches to multiple task-specific outputs. This eliminates the need to run separate inference processes for each task, significantly reducing power consumption while maintaining comprehensive task coverage across classification, detection, and segmentation.
4Reliability
If individual task-specific models are trained, then task specialization is improved, but training data requirements increase
Solution Approach 1:
The patent trains a single unified model on combined datasets from multiple tasks simultaneously, leveraging shared feature representations. This approach reduces the total amount of training data needed compared to training separate specialized models for each task, while the task-specific output layers ensure each task receives appropriate specialized processing during inference.
Data Source
AI summary
Systems and methods of the present disclosure are directed to a method for generating a machine-learned multitask model configured to perform tasks. The method can include obtaining a machine-learned multitask search model comprising candidate nodes. The method can include obtaining tasks and machine-learned task controller models associated with the tasks. As an example, for a task, the method can include using the task controller model to route a subset of the candidate nodes in a machine-learned task submodel for the corresponding task. The method can include inputting task input data to the task submodel to obtain a task output. The method can include generating, using the task output, a feedback value based on an objective function. The method can include adjusting parameters of the task controller model based on the feedback value.