Multi-Task Learning Network with CKA Layer Sharing for Lower Compute
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems require multiple neural networks for different tasks, consuming significant computing resources and time, which is inefficient for vehicles and other resource-constrained environments.
Innovation Solution
Implementing a multi-task learning (MTL) neural network that combines multiple tasks into a single network by sharing layers, using centered kernel alignment (CKA) to optimize layer redundancy and reduce resource consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple separate neural networks are used for different tasks, then task performance and accuracy are improved, but computing resource consumption and processing time increase significantly
Solution Approach 1:
The patent combines multiple separate neural networks into a single multi-task learning network that shares common layers and parameters across different tasks. This merging reduces the total number of parameters and computing operations required, directly addressing the contradiction by maintaining task performance while reducing computing resource consumption through shared feature extraction layers and task-specific head networks
Solution Approach 2:
The patent creates a universal neural network architecture that can perform multiple different tasks simultaneously through a shared backbone network. The multi-task learning framework enables a single network to handle various computer vision tasks (such as object detection, segmentation, and classification) by sharing common computational resources while maintaining task-specific capabilities through separate output heads
2Measurement precision
If multiple separate neural networks are used for different tasks, then task specialization is improved, but processing time and computational efficiency deteriorate
Solution Approach 1:
The patent merges multiple specialized neural networks into a single multi-task network that processes multiple tasks in parallel through shared computational layers. This combining approach maintains task specialization through separate task-specific output heads while improving processing efficiency by eliminating redundant computations and enabling joint optimization across all tasks
Solution Approach 2:
The patent enables continuous and simultaneous processing of multiple tasks through a unified neural network architecture. The shared backbone network continuously extracts features that are then distributed to multiple task-specific heads, allowing all tasks to be processed in parallel without the sequential delays that would occur with separate networks, thereby maintaining specialization while improving overall productivity
3Productivity
If a single multi-task neural network is used, then computing resource usage and processing time are reduced, but network complexity and training difficulty increase
Solution Approach 1:
The patent segments the multi-task neural network into distinct modular components: a shared backbone network for common feature extraction and task-specific head networks for individual task outputs. This segmentation reduces overall complexity by organizing the architecture into manageable, independent modules that can be trained and optimized separately while maintaining the efficiency benefits of a unified multi-task framework
4Measurement precision
If multiple separate neural networks are used, then task-specific optimization is improved, but resource consumption and time requirements worsen
Solution Approach 1:
The patent combines multiple task-specific neural networks into a single multi-task learning framework that shares common computational layers. This merging enables simultaneous training of all tasks in parallel, dramatically reducing the total training time compared to sequential training of separate networks, while maintaining task-specific accuracy through dedicated output heads and task-specific loss functions
Data Source
AI summary
A computer that includes a processor and a memory, the memory including instructions executable by the processor to execute a multi-task learning (MTL) neural network that includes multi-task layers and single-task layers, and that is configured to receive an image as input to a first one of the multi-task layers and to output respective predictions for each of a plurality of tasks from respective ones of the single-task layers. Activation tensors can be output from the multi-task layers and can be provided as input to the single-task layers.