Shared Backbone Multi-Task Deep Learning Model for Mobile Memory Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning systems require significant memory space, making them challenging to apply in mobile scenarios where memory is limited, and existing lightweight networks often exceed memory limits even when used independently.
Innovation Solution
A lightweight framework using a shared backbone network with multiple sub-networks for feature extraction, combined with a composite loss function for training, reduces model size and enhances efficiency for performing multiple deep-learning tasks like object detection and image classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If deep learning systems are used to perform multiple machine learning tasks, then task completion capability is improved, but memory space requirement increases
Solution Approach 1:
The patent implements a shared backbone network that serves multiple machine learning tasks simultaneously. The backbone network is trained to perform multiple tasks through multi-task learning, allowing a single model to handle various image processing tasks such as classification, detection, and segmentation without requiring separate models for each task.
Solution Approach 2:
The patent divides the overall model into two segments: a shared backbone network that extracts common features, and task-specific head networks that process features for individual tasks. This segmentation allows the backbone to be lightweight while enabling multiple tasks through specialized heads, reducing overall memory requirements compared to having complete separate models for each task.
2Quantity of substance
If lightweight networks are used to reduce model size, then memory space requirement decreases, but processing speed may be affected
Solution Approach 1:
The patent employs parameter sharing across multiple tasks in the backbone network, where the same parameters serve multiple functions. This parameter sharing reduces the total number of parameters compared to having separate models for each task, thereby reducing model size while maintaining processing efficiency through optimized feature extraction.
3Reliability
If multiple independent models are used for different tasks, then task specialization is improved, but device complexity increases
Solution Approach 1:
The patent merges multiple task-specific models into a single unified multi-task model with a shared backbone. Instead of managing separate independent models for each task, the system combines them into one architecture where the backbone shares features across tasks, reducing deployment complexity while maintaining task-specific capabilities through dedicated head networks.
Data Source
AI summary
A method, computer program, and computer system are provided for performing multiple machine learning tasks through a shared framework. Data corresponding to a plurality of predetermined machine learning tasks is received. One or more steps of the machine learning tasks associated with the received data is performed on the received data by a shared backbone of a machine learning model. The predetermined plurality of machine learning tasks is completed on the received data by a plurality of sub-networks associated with each of the plurality of predetermined machine learning tasks.