Distributed Neural Network Layer Segmentation for Thermal Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Head-mounted display systems face limitations in processing capabilities due to size and heat dissipation constraints, making it difficult to perform computationally heavy tasks locally, and offloading these tasks to external units can cause lag and energy inefficiency.
Innovation Solution
A method where a first device with a neural network processes input data, reduces it to a smaller dataset, and communicates this reduced data to a second device for further processing, allowing the first device to perform actions without transmitting the full dataset, thereby optimizing energy consumption and reducing data transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If computationally heavy processes are performed locally on the head mounted display processing circuitry, then processing speed and responsiveness are improved, but the device exceeds size and heat dissipation constraints
Solution Approach 1:
The neural network is divided into multiple layers distributed across different devices. The first device (head mounted display) executes first one or more layers, while the second device (external processing unit) executes second one or more layers. This segmentation allows computationally intensive operations to be offloaded to external devices, reducing local heat generation while maintaining processing speed.
2Temperature
If computationally heavy processes are off-loaded to an external processing unit, then heat dissipation and size constraints are satisfied, but data transmission energy consumption and lag increase
Solution Approach 1:
Only the necessary portion of data (input data or intermediate data from first layers) is extracted and transmitted to the second device for processing by subsequent layers. This minimizes data transmission volume and associated energy consumption while still achieving the computational offloading needed to satisfy thermal constraints.
Solution Approach 2:
The first device performs preliminary processing by executing the first one or more layers of the neural network locally before transmitting results to the second device. This preliminary action reduces the amount of data that needs to be transmitted and processed externally, thereby reducing transmission energy requirements.
3Measurement precision
If the full input data set is transmitted to the second device for processing, then processing accuracy is maintained, but data transmission time and energy consumption increase
Solution Approach 1:
Only the essential input data or intermediate features are extracted and transmitted to the second device, rather than transmitting the complete original data set. This extraction approach maintains sufficient information for accurate feature detection by the second one or more layers while significantly reducing transmission time.
4Measurement precision
If more neural network layers are implemented locally on the first device, then processing accuracy improves, but device complexity and computational load increase
Solution Approach 1:
The neural network is segmented into multiple layers distributed across the first device and second device. This segmentation reduces the complexity burden on the first device (head mounted display) while maintaining the overall accuracy of the complete neural network through collaborative processing across both devices.
Data Source
AI summary
Disclosed herein is a method for using a neural network across multiple devices. The method can include receiving, by a first device configured with a first one or more layers of a neural network, input data for processing via the neural network implemented across the first device and a second device. The method can include outputting, by the first one or more layers of the neural network implemented on the first device, a data set that is reduced in size relative to the input data while identifying one or more features of the input data for processing by a second one or more layers of the neural network. The method can include communicating, by the first device, the data set to the second device for processing via the second one or more layers of the neural network implemented on the second device.