Tensor Encoding With Upsampled Feature Maps for Lower Bit Consumption
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression technologies, particularly in collaborative intelligence architectures, face challenges in efficiently compressing and transmitting tensor data between edge devices and cloud systems, leading to high computational burden and costly processing power requirements.
Innovation Solution
A method and system for encoding and decoding tensors using basis vectors derived from upsampling and downsampling feature maps, allowing for efficient compression and transmission of tensor data across distributed systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If existing video compression technologies are used for tensor data, then compression is achieved, but computational burden and processing power requirements increase
Solution Approach 1:
The patent transforms tensor data into a video format by changing the data representation parameters, enabling the use of video compression standards (H.264, H.265, AV1) that are optimized for this format. This parameter transformation allows efficient compression of tensor data while reducing computational burden compared to applying general-purpose compression to raw tensor data
Solution Approach 2:
The patent replaces complex general-purpose tensor compression algorithms with established video compression mechanisms that have been optimized over decades. By substituting the compression approach with video encoding standards, the system achieves efficient compression with lower computational requirements
2Quantity of substance
If tensor data is compressed using conventional methods, then compression is achieved, but task performance becomes sensitive to bitrate variations
Solution Approach 1:
The patent performs preliminary actions by transforming the tensor data into video format before compression, and by deriving basis vectors from upsampled and downsampled feature maps. This preliminary transformation ensures that the compressed data maintains task performance resilience across varying bitrates, as the video format and basis vectors preserve critical information more effectively
3Adaptability or versatility
If collaborative intelligence architectures are implemented, then distributed processing is achieved, but transmission and processing costs increase
Solution Approach 1:
The patent changes the data format parameter to video, which enables efficient compression and transmission in collaborative intelligence architectures. This parameter change reduces the bandwidth requirements for transmitting tensor data between edge devices and cloud systems, thereby reducing transmission and processing costs while maintaining distributed processing capabilities
Data Source
AI summary
A system and method of encoding a tensor. The tensor includes a first set of feature maps and a second set of feature maps, a feature map in the first set having a first size, and a feature map in the second set having a second size lager than the first size. The method comprises upsampling the first set of feature maps; deriving basis vectors by performing a predetermined process on a tensor including the upsampled first set of feature maps and the second set of feature maps, and deriving coefficients for the tensor using the derived basis vectors to encode the tensor.