Tensor Compression Profiles for AI Split Point Negotiation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding systems, including wavelet-based, object-based, and block-based systems, are inadequate for effective compression and processing of point clouds.
Innovation Solution
Implementing tensor-based compression profiles for split point configurations, which allow endpoints to negotiate and apply compression schemes based on performance thresholds, tensor characteristics, and available bandwidth, to compress and decompress intermediate data tensors in AI processing models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If current video coding systems (wavelet-based, object-based, block-based) are used for point cloud compression, then existing compression methods can be applied, but compression effectiveness and processing adequacy are insufficient
Solution Approach 1:
The patent transforms point cloud data into tensor representations and applies tensor-based compression profiles that operate on different parameter spaces than traditional video coding systems. By changing the data representation parameters (from video frames to point cloud tensors) and compression parameters (using tensor-specific metrics and operations), the system achieves both improved compression effectiveness and adequate processing reliability simultaneously
Solution Approach 2:
The patent replaces the mechanical video coding compression mechanisms (wavelet transforms, block-based DCT, motion compensation) with tensor-based mathematical operations and compression profiles. This substitution enables more effective compression for point cloud data structures while maintaining processing adequacy through tensor-specific algorithms designed for geometric data
2Productivity
If compression is applied to intermediate data tensors in AI processing models, then bandwidth usage is reduced and transmission efficiency is improved, but compression accuracy and performance thresholds may be compromised
Solution Approach 1:
The patent implements dynamic compression profiles that adapt to different intermediate data tensors based on their characteristics, performance thresholds, and available bandwidth. The compression level and method are dynamically adjusted for each tensor, allowing high compression for less critical data while maintaining higher accuracy for tensors requiring strict performance thresholds, thus achieving both transmission efficiency and compression accuracy
Solution Approach 2:
The patent applies different compression quality levels to different parts of the AI processing pipeline based on local requirements. Each intermediate data tensor can have its own compression profile tailored to its specific characteristics and the performance requirements of downstream operations, ensuring that compression accuracy is maintained where needed while maximizing transmission efficiency where possible
3Productivity
If compression profiles are negotiated between endpoints based on performance thresholds and tensor characteristics, then compression optimization is improved and resource utilization is enhanced, but system complexity and negotiation overhead increase
Solution Approach 1:
The patent standardizes the negotiation parameters for compression profiles into a manageable set of tensor characteristics (shape, data type, performance thresholds) and compression profile parameters. By defining a standardized parameter interface for negotiation, the system achieves effective compression optimization while controlling system complexity through parameter standardization
Solution Approach 2:
The patent creates a universal compression profile negotiation mechanism that can handle multiple tensor types, AI processing models, and performance requirements through a single standardized framework. This universal approach allows the same negotiation protocol to optimize compression across diverse scenarios, reducing system complexity compared to having separate mechanisms for each case
Data Source
Figure 1A
Figure 1B
Figure 1C
AI summary
Systems, methods, and instrumentalities are disclosed associated with tensor-based compression profile(s) for split point configuration(s). Compression profile(s) may be selected for a (e.g., each) tensor for a candidate split point configuration to negotiate. Compression profiles may apply to intermediate data tensors to fulfil an application task (e.g., data processing task). The compression profiles may be negotiated between endpoints. Endpoints may determine a compression scheme to apply to a tensor based on a negotiated split point configuration. The determined compression scheme may be applied to a tensor. Data may be sent (e.g., via a bitstream) between endpoints. The data may include a compressed tensor and/or split point information.